Professional Documents
Culture Documents
Inside
Table of Contents
1. Introduction
2. Booting
i. From bootloader to kernel
ii. First steps in the kernel setup code
iii. Video mode initialization and transition to protected mode
iv. Transition to 64-bit mode
v. Kernel decompression
3. Initialization
i. First steps in the kernel
ii. Early interrupts handler
iii. Last preparations before the kernel entry point
iv. Kernel entry point
v. Continue architecture-specific boot-time initializations
vi. Architecture-specific initializations, again...
vii. End of the architecture-specific initializations, almost...
viii. Scheduler initialization
ix. RCU initialization
x. End of initialization
4. Interrupts
i. Introduction
ii. Start to dive into interrupts
iii. Interrupt handlers
iv. Initialization of non-early interrupt gates
v. Implementation of some exception handlers
vi. Handling Non-Maskable interrupts
vii. Dive into external hardware interrupts
viii. Initialization of external hardware interrupts structures
5. Memory management
i. Memblock
ii. Fixmaps and ioremap
6. vsyscalls and vdso
7. SMP
8. Concepts
i. Per-CPU variables
ii. Cpumasks
9. Data Structures in the Linux Kernel
i. Doubly linked list
ii. Radix tree
10. Theory
i. Paging
ii. Elf64
iii. CPUID
iv. MSR
11. Initial ram disk
i. initrd
12. Misc
i. Kernel building and instalation
ii. Write and Submit your first Linux kernel Patch
iii. Data types in the kernel
2
Linux Inside
Linux Inside
linux-insides
A series of posts about the linux kernel and its insides.
The goal is simple - to share my modest knowledge about the internals of the linux kernel and help people who are
interested in the linux kernel internals, and other low-level subject matter.
Questions/Suggestions: Feel free about any questions or suggestions by pinging me at twitter @0xAX, adding issue or
just drop me email.
Support
Support If you like linux-insides you can support me with:
LICENSE
Licensed BY-NC-SA Creative Commons.
Contributions
Feel free to create issues or create pull-requests if you find any issues or my English is poor.
Please read CONTRIBUTING.md before pushing any changes.
Author
@0xAX
Introduction
Linux Inside
Introduction
Linux Inside
Booting
Linux Inside
IP 0xfff0
CS selector 0xf000
CS base 0xffff0000
The processor starts working in real mode and we need to back up a little to understand memory segmentation in this
mode. Real mode is supported in all x86-compatible processors, from 8086 to modern Intel 64-bit CPUs. The 8086
processor had a 20-bit address bus, which means that it could work with 0-2^20 bytes address space (1 megabyte). But it
only has 16-bit registers, and with 16-bit registers the maximum address is 2^16 or 0xffff (64 kilobytes). Memory
segmentation is used to make use of all of the address space available. All memory is divided into small, fixed-size
segments of 65535 bytes, or 64 KB. Since we cannot address memory below 64 KB with 16 bit registers, an alternate
method to do it was devised. An address consists of two parts: the beginning address of the segment and the offset from
From bootloader to kernel
Linux Inside
the beginning of this segment. To get a physical address in memory, we need to multiply the segment part by 16 and add
the offset part:
For example if CS:IP is 0x2000:0x0010 , the corresponding physical address will be:
But if we take the biggest segment part and offset: 0xffff:0xffff , it will be:
which is 65519 bytes over first megabyte. Since only one megabyte is accessible in real mode, 0x10ffef becomes
0x00ffef with disabled A20.
Ok, now we know about real mode and memory addressing. Let's get back to register values after reset.
CS register consists of two parts: the visible segment selector and hidden base address. We know predefined CS base
0xffff0000:0xfff0
In this way starting address formed by adding the base address to the value in the EIP register:
We get 0xfffffff0 which is 4GB - 16 bytes. This point is the Reset vector. This is the memory location at which CPU
expects to find the first instruction to execute after reset. It contains a jump instruction which usually points to the BIOS
entry point. For example, if we look in coreboot source code, we will see it:
.section ".reset"
.code16
.globl reset_vector
reset_vector:
.byte 0xe9
.int _start - ( . + 2 )
...
We can see here the jump instruction opcode - 0xe9 to the address _start - ( . + 2) . And we can see that reset section
is 16 bytes and starts at 0xfffffff0 :
SECTIONS {
_ROMTOP = 0xfffffff0;
. = _ROMTOP;
.reset . : {
*(.reset)
. = 15 ;
Linux Inside
BYTE(0x00);
}
}
Now the BIOS has started to work. After initializing and checking the hardware, it needs to find a bootable device. A boot
order is stored in the BIOS configuration. The function of boot order is to control which devices the kernel attempts to boot.
In the case of attempting to boot a hard drive, the BIOS tries to find a boot sector. On hard drives partitioned with an MBR
partition layout, the boot sector is stored in the first 446 bytes of the first sector (512 bytes). The final two bytes of the first
sector are 0x55 and 0xaa which signals the BIOS that the device is bootable. For example:
;
; Note: this example is written in Intel Assembly syntax
;
[BITS 16]
[ORG 0x7c00]
boot:
mov al, '!'
mov ah, 0x0e
mov bh, 0x00
mov bl, 0x07
int 0x10
jmp $
times 510-($-$$) db 0
db 0x55
db 0xaa
This will instruct QEMU to use the boot binary we just built as a disk image. Since the binary generated by the assembly
code above fulfills the requirements of the boot sector (the origin is set to 0x7c00 , and we end with the magic sequence).
QEMU will treat the binary as the master boot record(MBR) of a disk image.
We will see:
Linux Inside
In this example we can see that this code will be executed in 16 bit real mode and will start at 0x7c00 in memory. After the
start it calls the 0x10 interrupt which just prints ! symbol. It fills rest of 510 bytes with zeros and finish with two magic bytes
0xaa and 0x55 .
A real-world boot sector has code for continuing the boot process and the partition table instead of a bunch of 0's and an
exclamation point :) Ok so, from this point onwards BIOS hands over the control to the bootloader and we can go ahead.
NOTE: As you can read above the CPU is in real mode. In real mode, calculating the physical address in memory is done
as following:
Same as I mentioned before. But we have only 16 bit general purpose registers. The maximum value of 16 bit register is:
0xffff ; So if we take the biggest values the result will be:
Where 0x10ffef is equal to 1MB + 64KB - 16b . But a 8086 processor, which was the first processor with real mode. It had
20 bit address line and 2^20 = 1048576.0 is 1MB. So, it means that the actual memory available is 1MB.
General real mode's memory map is:
10
Linux Inside
But stop, at the beginning of post I wrote that first instruction executed by the CPU is located at the address 0xFFFFFFF0 ,
which is much bigger than 0xFFFFF (1MB). How can CPU access it in real mode? As I write about it and you can read in
coreboot documentation:
Bootloader
There are a number of bootloaders which can boot Linux, such as GRUB 2 and syslinux. The Linux kernel has a Boot
protocol which specifies the requirements for bootloaders to implement Linux support. This example will describe GRUB 2.
Now that the BIOS has chosen a boot device and transferred control to the boot sector code, execution starts from
boot.img. This code is very simple due to the limited amount of space available, and contains a pointer that it uses to jump
to the location of GRUB 2's core image. The core image begins with diskboot.img, which is usually stored immediately after
the first sector in the unused space before the first partition. The above code loads the rest of the core image into memory,
which contains GRUB 2's kernel and drivers for handling filesystems. After loading the rest of the core image, it executes
grub_main.
grub_main initializes console, gets base address for modules, sets root device, loads/parses grub configuration file, loads
modules etc. At the end of execution, grub_main moves grub to normal mode. grub_normal_execute (from grubcore/normal/main.c ) completes last preparation and shows a menu for selecting an operating system. When we select one
of grub menu entries, grub_menu_execute_entry begins to be executed, which executes grub boot command. It starts to
boot the selected operating system.
As we can read in the kernel boot protocol, the bootloader must read and fill some fields of kernel setup header which
starts at 0x01f1 offset from the kernel setup code. Kernel header arch/x86/boot/header.S starts from:
.globl hdr
hdr:
setup_sects: .byte 0
root_flags: .word ROOT_RDONLY
syssize: .long 0
ram_size: .word 0
vid_mode: .word SVGA_MODE
root_dev: .word 0
boot_flag: .word 0xAA55
The bootloader must fill this and the rest of the headers (only marked as write in the Linux boot protocol, for example this)
with values which it either got from command line or calculated. We will not see description and explanation of all fields of
kernel setup header, we will get back to it when kernel uses it. Anyway, you can find description of any field in the boot
protocol.
As we can see in kernel boot protocol, the memory map will be the following after kernel loading:
11
Linux Inside
| Protected-mode kernel |
100000 +------------------------+
| I/O memory hole |
0A0000 +------------------------+
| Reserved for BIOS | Leave as much as possible unused
~ ~
| Command line | (Can also be below the X+10000 mark)
X+10000 +------------------------+
| Stack/heap | For use by the kernel real-mode code.
X+08000 +------------------------+
| Kernel setup | The kernel real-mode code.
| Kernel boot sector | The kernel legacy boot sector.
X +------------------------+
| Boot loader |
So after the bootloader transferred control to the kernel, it starts somewhere at:
0x1000 + X + sizeof(KernelBootSector) + 1
where X is the address of kernel bootsector loaded. In my case X is 0x10000 , we can see it in memory dump:
Ok, now the bootloader has loaded Linux kernel into the memory, filled header fields and jumped to it. Now we can move
directly to the kernel setup code.
qemu-system-x86_64 vmlinuz-3.18-generic
12
Linux Inside
Actually header.S starts from MZ (see image above), error message printing and following PE header:
#ifdef CONFIG_EFI_STUB
# "MZ", MS-DOS header
.byte 0x4d
.byte 0x5a
#endif
...
...
...
pe_header:
.ascii "PE"
.word 0
It needs this for loading the operating system with UEFI. Here we will not see how it works (we will these later in the next
parts).
So the actual kernel setup entry point is:
Bootloader (grub2 and others) knows about this point ( 0x200 offset from MZ ) and makes a jump directly to this point,
despite the fact that header.S starts from .bstext section which prints error message:
//
// arch/x86/boot/setup.ld
//
. = 0; // current position
.bstext : { *(.bstext) } // put .bstext section to position 0
.bsdata : { *(.bsdata) }
13
Linux Inside
.globl _start
_start:
.byte 0xeb
.byte start_of_setup-1f
1:
//
// rest of the header
//
Here we can see jmp instruction opcode - 0xeb to the start_of_setup-1f point. Nf notation means following: 2f refers to
the next local 2: label. In our case it is label 1 which goes right after jump. It contains rest of setup header and right after
setup header we can see .entrytext section which starts at start_of_setup label.
Actually it's the first code which starts to execute besides previous jump instruction. After kernel setup got the control from
bootloader, first jmp instruction is located at 0x200 (first 512 bytes) offset from the start of kernel real mode. This we can
read in Linux kernel boot protocol and also see in grub2 source code:
It means that segment registers will have following values after kernel setup starts to work:
fs = es = ds = ss = 0x1000
cs = 0x1020
As I wrote above, grub2 loads kernel setup code at 0x10000 address and cs at 0x1020 because execution doesn't start
from the start of file, but from:
_start:
.byte 0xeb
.byte start_of_setup-1f
14
Linux Inside
jump , which is 512 bytes offset from the 4d 5a. Also need to align cs from 0x10200 to 0x10000 as all other segment
pushw %ds
pushw $6f
lretw
push ds value to stack, and address of 6 label and execute lretw instruction. When we call lretw , it loads address of
label 6 to instruction pointer register and cs with value of ds . After it we will have ds and cs with the same values.
Stack Setup
Actually, almost all of the setup code is preparation for C language environment in the real mode. The next step is checking
of ss register value and making of correct stack if ss is wrong:
Here we can see aligning of dx (contains sp given by bootloader) to 4 bytes and checking that it is not zero. If it is zero
we put 0xfffc (4 byte aligned address before maximum segment size - 64 KB) to dx . If it is not zero we continue to use
sp given by bootloader (0xf7f4 in my case). After this we put ax value to ss which stores correct segment address
0x10000 and set up correct sp . After it we have correct stack:
15
Linux Inside
1. In the second case ( ss != ds ), first of all put _end (address of end of setup code) value in dx . And check loadflags
header field with testb instruction too see if we can use heap or not. loadflags is a bitmask header which is defined
as:
If CAN_USE_HEAP bit is set, put heap_end_ptr to dx which points to _end and add STACK_SIZE (minimal stack size - 512
bytes) to it. After this if dx is not carry, jump to 2 (it will not be carry, dx = _end + 512) label as in previous case and make
correct stack.
1. The last case when CAN_USE_HEAP is not set, we just use minimal stack from _end to _end + STACK_SIZE :
16
Linux Inside
BSS Setup
The last two steps that need to happen before we can jump to the main C code, are that we need to set up the BSS area,
and check the "magic" signature. Firstly, signature checking:
This simply consists of comparing the setup_sig against the magic number 0x5a5aaa55 . If they are not equal, a fatal error is
reported.
But if the magic number matches, knowing we have a set of correct segment registers, and a stack, we need only setup the
BSS section before jumping into the C code.
The BSS section is used for storing statically allocated, uninitialized, data. Linux carefully ensures this area of memory is
first blanked, using the following code:
First of all the __bss_start address is moved into di , and the _end + 3 address (+3 - aligns to 4 bytes) is moved into cx .
The eax register is cleared (using an xor instruction), and the bss section size ( cx - di ) is calculated and put into cx .
Then, cx is divided by four (the size of a 'word'), and the stosl instruction is repeatedly used, storing the value of eax
(zero) into the address pointed to by di , and automatically increasing di by four (this occurs until cx reaches zero). The
net effect of this code, is that zeros are written through all words in memory from __bss_start to _end :
17
Linux Inside
Jump to main
That's all, we have the stack, BSS and now we can jump to the main() C function:
calll main
The main() function is located in arch/x86/boot/main.c. What will be there? We will see it in the next part.
Conclusion
This is the end of the first part about Linux kernel internals. If you have questions or suggestions, ping me in twitter 0xAX,
drop me email or just create issue. In the next part we will see first C code which executes in Linux kernel setup,
implementation of memory routines as memset , memcpy , earlyprintk implementation and early console initialization and
many more.
Please note that English is not my first language and I am really sorry for any inconvenience. If you found any
mistakes please send me PR to linux-internals.
Links
Intel 80386 programmer's reference manual 1986
Minimal Boot Loader for Intel Architecture
8086
80386
Reset vector
Real mode
Linux kernel boot protocol
CoreBoot developer manual
Ralf Brown's Interrupt List
Power supply
Power good signal
18
Linux Inside
Protected mode
Before we can move to the native Intel64 Long Mode, the kernel must switch the CPU into protected mode.
What is protected mode? Protected mode was first added to the x86 architecture in 1982 and was the main mode of Intel
processors from the 80286 processor until Intel 64 and long mode came.
The main reason to move away from Real mode is that there is very limited access to the RAM. As you may remember
from the previous part, there is only 220 bytes or 1 Megabyte, sometimes even only 640 Kilobytes of RAM available in the
Real mode.
Protected mode brought many changes, but the main one is the difference in memory management. The 20-bit address
bus was replaced with a 32-bit address bus. It allowed access to 4 Gigabytes of memory vs 1 Megabyte of real mode. Also
paging support was added, which you can read about in the next sections.
Memory management in Protected mode is divided into two, almost independent parts:
Segmentation
Paging
Here we will only see segmentation. Paging will be discussed in the next sections.
As you can read in the previous part, addresses consist of two parts in real mode:
Base address of the segment
Offset from the segment base
And we can get the physical address if we know these two parts by:
Memory segmentation was completely redone in protected mode. There are no 64 Kilobyte fixed-size segments. Instead,
the size and location of each segment is described by an associated data structure called Segment Descriptor. The
First steps in the kernel setup code
19
Linux Inside
segment descriptors are stored in a data structure called Global Descriptor Table (GDT).
The GDT is a structure which resides in memory. It has no fixed place in the memory so, its address is stored in the special
GDTR register. Later we will see the GDT loading in the Linux kernel code. There will be an operation for loading it into
lgdt gdt
where the lgdt instruction loads the base address and limit(size) of global descriptor table to the GDTR register. GDTR is a
48-bit register and consists of two parts:
size(16-bit) of global descriptor table;
address(32-bit) of the global descriptor table.
As mentioned above the GDT contains segment descriptors which describe memory segments. Each descriptor is 64-bits
in size. The general scheme of a descriptor is:
31 24 19 16 7 0
-----------------------------------------------------------| | |B| |A| | | | |0|E|W|A| |
| BASE 31:24 |G|/|L|V| LIMIT |P|DPL|S| TYPE | BASE 23:16 | 4
| | |D| |L| 19:16 | | | |1|C|R|A| |
-----------------------------------------------------------| | |
| BASE 15:0 | LIMIT 15:0 | 0
| | |
------------------------------------------------------------
Don't worry, I know it looks a little scary after real mode, but it's easy. For example LIMIT 15:0 means that bit 0-15 of the
Descriptor contain the value for the limit. The rest of it is in LIMIT 16:19. So, the size of Limit is 0-19 i.e 20-bits. Let's take a
closer look at it:
1. Limit[20-bits] is at 0-15,16-19 bits. It defines length_of_segment - 1 . It depends on G (Granularity) bit.
if G (bit 55) is 0 and segment limit is 0, the size of the segment is 1 Byte
if G is 1 and segment limit is 0, the size of the segment is 4096 Bytes
if G is 0 and segment limit is 0xfffff, the size of the segment is 1 Megabyte
if G is 1 and segment limit is 0xfffff, the size of the segment is 4 Gigabytes
So, it means that if
if G is 0, Limit is interpreted in terms of 1 Byte and the maximum size of the segment can be 1 Megabyte.
if G is 1, Limit is interpreted in terms of 4096 Bytes = 4 KBytes = 1 Page and the maximum size of the segment
can be 4 Gigabytes. Actually when G is 1, the value of Limit is shifted to the left by 12 bits. So, 20 bits + 12 bits =
32 bits and 232 = 4 Gigabytes.
2. Base[32-bits] is at (0-15, 32-39 and 56-63 bits). It defines the physical address of the segment's starting location.
3. Type/Attribute (40-47 bits) defines the type of segment and kinds of access to it.
S flag at bit 44 specifies descriptor type. If S is 0 then this segment is a system segment, whereas if S is 1 then
this is a code or data segment (Stack segments are data segments which must be read/write segments).
To determine if the segment is a code or data segment we can check its Ex(bit 43) Attribute marked as 0 in the above
diagram. If it is 0, then the segment is a Data segment otherwise it is a code segment.
A segment can be of one of the following types:
20
Linux Inside
As we can see the first bit(bit 43) is 0 for a data segment and 1 for a code segment. The next three bits(40, 41, 42, 43)
are either EWA (Expansion Writable Accessible) or CRA(Conforming Readable Accessible).
if E(bit 42) is 0, expand up other wise expand down. Read more here.
if W(bit 41)(for Data Segments) is 1, write access is allowed otherwise not. Note that read access is always allowed on
data segments.
A(bit 40) - Whether the segment is accessed by processor or not.
C(bit 43) is conforming bit(for code selectors). If C is 1, the segment code can be executed from a lower level privilege
for e.g user level. If C is 0, it can only be executed from the same privilege level.
R(bit 41)(for code segments). If 1 read access to segment is allowed otherwise not. Write access is never allowed to
code segments.
1. DPL[2-bits] (Descriptor Privilege Level) is at bits 45-46. It defines the privilege level of the segment. It can be 0-3
where 0 is the most privileged.
2. P flag(bit 47) - indicates if the segment is present in memory or not. If P is 0, the segment will be presented as invalid
and the processor will refuse to read this segment.
3. AVL flag(bit 52) - Available and reserved bits. It is ignored in Linux.
4. L flag(bit 53) - indicates whether a code segment contains native 64-bit code. If 1 then the code segment executes in
64 bit mode.
5. D/B flag(bit 54) - Default/Big flag represents the operand size i.e 16/32 bits. If it is set then 32 bit otherwise 16.
Segment registers don't contain the base address of the segment as in real mode. Instead they contain a special structure Segment Selector . Each Segment Descriptor has an associated Segment Selector. Segment Selector is a 16-bit structure:
Where,
Index shows the index number of the descriptor in the GDT.
TI(Table Indicator) shows where to search for the descriptor. If it is 0 then search in the Global Descriptor Table(GDT)
otherwise it will look in Local Descriptor Table(LDT).
First steps in the kernel setup code
21
Linux Inside
The algorithm for the transition from real mode into protected mode is:
Disable interrupts
Describe and load GDT with lgdt instruction
Set PE (Protection Enable) bit in CR0 (Control Register 0)
Jump to protected mode code
We will see the complete transition to protected mode in the linux kernel in the next part, but before we can move to
First steps in the kernel setup code
22
Linux Inside
GLOBAL(memcpy)
pushw %si
pushw %di
movw %ax, %di
movw %dx, %si
pushw %cx
shrw $2, %cx
rep; movsl
popw %cx
andw $3, %cx
rep; movsb
popw %di
popw %si
retl
ENDPROC(memcpy)
Yeah, we just moved to C code and now assembly again :) First of all we can see that memcpy and other routines which are
defined here, start and end with the two macros: GLOBAL and ENDPROC . GLOBAL is described in
arch/x86/include/asm/linkage.h which defines globl directive and the label for it. ENDPROC is described in
include/linux/linkage.h which marks name symbol as function name and ends with the size of the name symbol.
Implementation of memcpy is easy. At first, it pushes values from si and di registers to the stack because their values will
change during the memcpy , so it pushes them on the stack to preserve their values. memcpy (and other functions in copy.S)
use fastcall calling conventions. So it gets its incoming parameters from the ax , dx and cx registers. Calling memcpy
looks like this:
So,
ax will contain the address of the boot_params.hdr in bytes
dx will contain the address of hdr in bytes
cx will contain the size of hdr in bytes.
memcpy puts the address of boot_params.hdr into si and saves the size on the stack. After this it shifts to the right on 2
size (or divide on 4) and copies from si to di by 4 bytes. After this we restore the size of hdr again, align it by 4 bytes
First steps in the kernel setup code
23
Linux Inside
and copy the rest of the bytes from si to di byte by byte (if there is more). Restore si and di values from the stack in
the end and after this copying is finished.
Console initialization
After the hdr is copied into boot_params.hdr , the next step is console initialization by calling the console_init function
which is defined in arch/x86/boot/early_serial_console.c.
It tries to find the earlyprintk option in the command line and if the search was successful, it parses the port address and
baud rate of the serial port and initializes the serial port. Value of earlyprintk command line option can be one of the:
* serial,0x3f8,115200
* serial,ttyS0,115200
* ttyS0,115200
if (cmdline_find_option_bool("debug"))
puts("early console in setup code\n");
The definition of puts is in tty.c. As we can see it prints character by character in a loop by calling the putchar function.
Let's look into the putchar implementation:
__attribute__((section(".inittext"))) means that this code will be in the .inittext section. We can find it in the linker file
setup.ld.
First of all, put_char checks for the \n symbol and if it is found, prints \r before. After that it outputs the character on the
VGA screen by calling the BIOS with the 0x10 interrupt call:
Here initregs takes the biosregs structure and first fills biosregs with zeros using the memset function and then fills it
with register values.
24
Linux Inside
GLOBAL(memset)
pushw %di
movw %ax, %di
movzbl %dl, %eax
imull $0x01010101,%eax
pushw %cx
shrw $2, %cx
rep; stosl
popw %cx
andw $3, %cx
rep; stosb
popw %di
retl
ENDPROC(memset)
As you can read above, it uses the fastcall calling conventions like the memcpy function, which means that the function
gets parameters from ax , dx and cx registers.
Generally memset is like a memcpy implementation. It saves the value of the di register on the stack and puts the ax
value into di which is the address of the biosregs structure. Next is the movzbl instruction, which copies the dl value to
the low 2 bytes of the eax register. The remaining 2 high bytes of eax will be filled with zeros.
The next instruction multiplies eax with 0x01010101 . It needs to because memset will copy 4 bytes at the same time. For
example, we need to fill a structure with 0x7 with memset. eax will contain 0x00000007 value in this case. So if we multiply
eax with 0x01010101 , we will get 0x07070707 and now we can copy these 4 bytes into the structure. memset uses rep;
stosl instructions for copying eax into es:di .
The rest of the memset function does almost the same as memcpy .
After that biosregs structure is filled with memset , bios_putchar calls the 0x10 interrupt which prints a character.
Afterwards it checks if the serial port was initialized or not and writes a character there with serial_putchar and inb/outb
instructions if it was set.
Heap initialization
After the stack and bss section were prepared in header.S (see previous part), the kernel needs to initialize the heap with
the init_heap function.
First of all init_heap checks the CAN_USE_HEAP flag from the loadflags in the kernel setup header and calculates the end of
the stack if this flag was set:
char *stack_end;
if (boot_params.hdr.loadflags & CAN_USE_HEAP) {
asm("leal %P1(%%esp),%0"
: "=r" (stack_end) : "i" (-STACK_SIZE));
25
Linux Inside
which means heap_end_ptr or _end + 512 ( 0x200h ). And at the last is checked that whether heap_end is greater than
stack_end . If it is then stack_end is assigned to heap_end to make them equal.
Now the heap is initialized and we can use it using the GET_HEAP method. We will see how it is used, how to use it and how
the it is implemented in the next posts.
CPU validation
The next step as we can see is cpu validation by validate_cpu from arch/x86/boot/cpu.c.
It calls the check_cpu function and passes cpu level and required cpu level to it and checks that the kernel launches on the
right cpu level.
check_cpu checks the cpu's flags, presence of long mode in case of x86_64(64-bit) CPU, checks the processor's vendor
and makes preparation for certain vendors like turning off SSE+SSE2 for AMD if they are missing, etc.
Memory detection
The next step is memory detection by the detect_memory function. detect_memory basically provides a map of available
RAM to the cpu. It uses different programming interfaces for memory detection like 0xe820 , 0xe801 and 0x88 . We will see
only the implementation of 0xE820 here.
Let's look into the detect_memory_e820 implementation from the arch/x86/boot/memory.c source file. First of all, the
detect_memory_e820 function initializes the biosregs structure as we saw above and fills registers with special values for the
0xe820 call:
initregs(&ireg);
ireg.ax = 0xe820;
ireg.cx = sizeof buf;
ireg.edx = SMAP;
ireg.di = (size_t)&buf;
Next is a loop where data about the memory will be collected. It starts from the call of the 0x15 BIOS interrupt, which writes
one line from the address allocation table. For getting the next line we need to call this interrupt again (which we do in the
loop). Before the next call ebx must contain the value returned previously:
26
Linux Inside
Ultimately, it does iterations in the loop to collect data from the address allocation table and writes this data into the
e820entry array:
Keyboard initialization
The next step is the initialization of the keyboard with the call of the keyboard_init() function. At first keyboard_init
initializes registers using the initregs function and calling the 0x16 interrupt for getting the keyboard status.
initregs(&ireg);
ireg.ah = 0x02; /* Get keyboard status */
intcall(0x16, &ireg, &oreg);
boot_params.kbd_status = oreg.al;
After this it calls 0x16 again to set repeat rate and delay.
Querying
The next couple of steps are queries for different parameters. We will not dive into details about these queries, but will get
back to it in later parts. Let's take a short look at these functions:
The query_mca routine calls the 0x15 BIOS interrupt to get the machine model number, sub-model number, BIOS revision
level, and other hardware-specific attributes:
int query_mca(void)
{
struct biosregs ireg, oreg;
u16 len;
initregs(&ireg);
ireg.ah = 0xc0;
intcall(0x15, &ireg, &oreg);
if (oreg.eflags & X86_EFLAGS_CF)
27
Linux Inside
It fills the ah register with 0xc0 and calls the 0x15 BIOS interruption. After the interrupt execution it checks the carry flag
and if it is set to 1, the BIOS doesn't support (MCA)[https://en.wikipedia.org/wiki/Micro_Channel_architecture]. If carry flag
is set to 0, ES:BX will contain a pointer to the system information table, which looks like this:
Next we call the set_fs routine and pass the value of the es register to it. Implementation of set_fs is pretty simple:
This function contains inline assembly which gets the value of the seg parameter and puts it into the fs register. There
are many functions in boot.h like set_fs , for example set_gs , fs , gs for reading a value in it etc...
At the end of query_mca it just copies the table which pointed to by es:bx to the boot_params.sys_desc_table .
The next step is getting Intel SpeedStep information by calling the query_ist function. First of all it checks the CPU level
and if it is correct, calls 0x15 for getting info and saves the result to boot_params .
The following query_apm_bios function gets Advanced Power Management information from the BIOS. query_apm_bios
calls the 0x15 BIOS interruption too, but with ah = 0x53 to check APM installation. After the 0x15 execution,
query_apm_bios functions checks PM signature (it must be 0x504d ), carry flag (it must be 0 if APM supported) and value of
28
Linux Inside
mode interface. In the end it fills boot_params.apm_bios_info with values obtained from the BIOS.
Note that query_apm_bios will be executed only if CONFIG_APM or CONFIG_APM_MODULE was set in configuration file:
The last is the query_edd function, which queries Enhanced Disk Drive information from the BIOS. Let's look into the
query_edd implementation.
First of all it reads the edd option from kernel's command line and if it was set to off then query_edd just returns.
If EDD is enabled, query_edd goes over BIOS-supported hard disks and queries EDD information in the following loop:
where 0x80 is the first hard drive and the value of EDD_MBR_SIG_MAX macro is 16. It collects data into the array of edd_info
structures. get_edd_info checks that EDD is present by invoking the 0x13 interrupt with ah as 0x41 and if EDD is
present, get_edd_info again calls the 0x13 interrupt, but with ah as 0x48 and si containing the address of the buffer
where EDD information will be stored.
Conclusion
This is the end of the second part about Linux kernel internals. In the next part we will see video mode setting and the rest
of preparations before transition to protected mode and directly transitioning into it.
If you have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you found any
mistakes please send me a PR to linux-internals.
Links
Protected mode
Protected mode
Long mode
Nice explanation of CPU Modes with code
How to Use Expand Down Segments on Intel 386 and Later CPUs
earlyprintk documentation
Kernel Parameters
Serial console
Intel SpeedStep
APM
EDD specification
29
Linux Inside
30
Linux Inside
switching into the protected mode and transition into it in this part.
NOTE If you don't know anything about protected mode, you can find some information about it in the previous part. Also
there are a couple of links which can help you.
As i wrote above, we will start from the set_video function which defined in the arch/x86/boot/video.c source code file. We
can see that it starts with getting of video mode from the boot_params.hdr structure:
which we filled in the copy_boot_params function (you can read about it in the previous post). vid_mode is an obligatory field
which filled by the bootloader. You can find information about it in the kernel boot protocol:
vga=<mode>
<mode> here is either an integer (in C notation, either
decimal, octal, or hexadecimal) or one of the strings
"normal" (meaning 0xFFFF), "ext" (meaning 0xFFFE) or "ask"
(meaning 0xFFFD). This value should be entered into the
vid_mode field, as it is used by the kernel before the command
line is parsed.
So we can add vga option to the grub or another bootloader configuration file and it will pass this option to the kernel
command line. This option can have different values as we can read from the description, for example it can be integer
number or ask . If you will pass ask , you see menu like this:
31
Linux Inside
which will suggest to select video mode. We will look on it's implementation, but before we must to know about another
things.
If you will read source code of the kernel, you'll see it very often, so it will be good to remember about it.
Heap API
As we got vid_mode from the boot_params.hdr , we can see call of the RESET_HEAP in the set_video function. RESET_HEAP is
a macro which defined in the boot.h and looks as:
If you read second part, you can remember that we initialized the heap with the init_heap function. Since we can use heap,
we have a couple functions for it which defined in the boot.h . They are:
#define RESET_HEAP()...
As we saw just now. It uses for resetting the heap by setting the HEAP variable equal to _end , where _end is just:
32
Linux Inside
#define GET_HEAP(type, n) \
((type *)__get_heap(sizeof(type),__alignof__(type),(n)))
Let's try to understand how GET_HEAP works. We can see here that HEAP (which equal to _end after RESET_HEAP() ) is the
address of aligned memory according to a parameter. After it we save memory address from HEAP to the tmp variable,
move HEAP to the end of allocated block and return tmp which is start address of allocated memory.
And the last function is:
which subtracts value of the HEAP from the heap_end (we calculated it in the previous part) and returns 1 if there is enough
memory for n .
That's all. Now we have simple API for heap and can setup video mode.
33
Linux Inside
0x10 BIOS interruption. After interruption successfully executed, it returns row and column in the DL and DH registers.
Row and column will be stored in the orig_x and orig_y fields from the the boot_params.screen_info structure. After
store_cursor_position executed, store_video_mode function will be called. It just gets current video mode and stores it in
the boot_params.screen_info.orig_video_mode .
After this, it checks current video mode and set the video_segment . After the BIOS transfers control to the boot sector, the
following addresses are video memory:
So we set the video_segment variable to 0xb000 if current video mode is MDA, HGC, VGA in monochrome mode or 0xb800
in color mode. After setup of the address of the video segment need to store font size in the
boot_params.screen_info.orig_video_points with:
set_fs(0);
font_size = rdfs16(0x485);
boot_params.screen_info.orig_video_points = font_size;
First of all we put 0 to the FS register with set_fs function. We already saw functions like set_fs in the previous part.
They are all defined in the boot.h. Next we read value which located at address 0x485 (this memory location used to get
the font size) and save font size in the boot_params.screen_info.orig_video_points .
The next we get amount of columns and rows by address 0x44a and stores they in the
boot_params.screen_info.orig_video_cols and boot_params.screen_info.orig_video_lines . After this, execution of the
store_mode_params is finished.
The next we can see save_screen function which just saves screen content to the heap. This function collects all data
which we got in the previous functions like rows and columns amount and etc... to the saved_screen structure, which
defined as:
if (!heap_free(saved.x*saved.y*sizeof(u16)+512))
return;
and allocates space in the heap if it is enough and stores saved_screen in it.
The next call is probe_cards(0) from the arch/x86/boot/video-mode.c. It goes over all video_cards and collects number of
modes provided by the cards. Here is the interesting moment, we can see the loop:
34
Linux Inside
but video_cards not declared anywhere. Answer is simple: Every video mode presented in the x86 kernel setup code has
definition like this:
struct card_info {
const char *card_name;
int (*set_mode)(struct mode_info *mode);
int (*probe)(void);
struct mode_info *modes;
int nmodes;
int unsafe;
u16 xmode_first;
u16 xmode_n;
};
is in the .videocards segment. Let's look on the arch/x86/boot/setup.ld linker file, we can see there:
.videocards : {
video_cards = .;
*(.videocards)
video_cards_end = .;
}
It means that video_cards is just memory address and all card_info structures are placed in this segment. It means that all
card_info structures are placed between video_cards and video_cards_end , so we can use it in a loop to go over all of it.
After probe_cards executed we have all structures like static __videocard video_vga with filled nmodes (number of video
modes).
After that probe_cards executed, we move to the main loop in the setup_video function. There is infinite loop which tries to
setup video mode with the set_mode function or prints menu if we passed vid_mode=ask to the kernel command line or
video mode is undefined.
The set_mode function is defined in the video-mode.c and gets only one parameter - mode which is number of video mode
(we got it or from the menu or in the start of the setup_video , from kernel setup header).
set_mode function checks the mode and calls raw_set_mode function. The raw_set_mode calls set_mode function for selected
card. We can get access to this function from the card_info structure, every video mode defines this structure with filled
value which depends on video mode (for example for vga it is video_vga.set_mode function, see above example of
card_info structure for vga ). video_vga.set_mode is vga_set_mode , which checks vga mode and call function depend on
mode:
35
Linux Inside
force_x = mode->x;
force_y = mode->y;
switch (mode->mode) {
case VIDEO_80x25:
break;
case VIDEO_8POINT:
vga_set_8font();
break;
case VIDEO_80x43:
vga_set_80x43();
break;
case VIDEO_80x28:
vga_set_14font();
break;
case VIDEO_80x30:
vga_set_80x30();
break;
case VIDEO_80x34:
vga_set_80x34();
break;
case VIDEO_80x60:
vga_set_80x60();
break;
}
return 0;
}
Every function which setups video mode, just call 0x10 BIOS interruption with certain value in the AH register. After this we
have set video mode and now we can switch to the protected mode.
go_to_protected_mode defined in the arch/x86/boot/pm.c. It contains some functions which make last preparations before
we can jump into protected mode, so let's look on it and try to understand what they do and how it works.
At first we see call of realmode_switch_hook function in the go_to_protected_mode . This function invokes real mode switch
hook if it is present and disables NMI. Hooks are used if bootloader runs in a hostile environment. More about hooks you
can read in the boot protocol (see ADVANCED BOOT LOADER HOOKS). readlmode_swtich hook presents pointer to the
16-bit real mode far subroutine which disables non-maskable interruptions. After we checked realmode_switch hook (it
doesn't present for me), there is disabling of non-maskable interruptions:
asm volatile("cli");
outb(0x80, 0x70);
io_delay();
At first there is inline assembly instruction with cli instruction which clears the interrupt flag ( IF ), after this external
interrupts disabled. Next line disables NMI (non-maskable interruption). Interruption is a signal to the CPU which emitted by
hardware or software. After getting signal, CPU stops to execute current instructions sequence and transfers control to the
interruption handler. After interruption handler finished it's work, it transfers control to the interrupted instruction. Nonmaskable interruptions (NMI) - interruptions which are always processed, independently of permission. We will not dive into
details interruptions now, but will back to it in the next posts.
Let's back to the code. We can see that second line is writing 0x80 (disabled bit) byte to the 0x70 (CMOS Address
register). And call the io_delay function after it. io_delay which initiates small delay and looks like:
36
Linux Inside
Outputting any byte to the port 0x80 should delay exactly 1 microsecond. Sow we can write any value (value from AL
register in our case) to the 0x80 port. After this delay realmode_switch_hook function finished execution and we can move to
the next function.
The next function is enable_a20 , which enables A20 line. This function defined in the arch/x86/boot/a20.c and it tries to
enable A20 gate with different methods. The first is a20_test_short function which checks is A20 already enabled or not
with a20_test function:
First of all we put 0x0000 to the FS register and 0xffff to the GS register. Next we read value by address A20_TEST_ADDR
(it is 0x200 ) and put this value into saved variable and ctr . Next we write updated ctr value into fs:gs with wrfs32
function, make little delay, and read value into the GS register by address A20_TEST_ADDR+0x10 , if it's not zero we already
have enabled a20 line. If A20 is disabled, we try to enabled it with different method which you can find in the a20.c . For
example with call of 0x15 BIOS interruption with AH=0x2041 and etc... If enabled_a20 function finished with fail, printed error
message and called function die . You can remember it from the first source code file where we started arch/x86/boot/header.S:
die:
hlt
jmp die
.size die, .-die
After the a20 gate successfully enabled, there are reset coprocessor and mask all interrupts. And after all of this
preparations, we can see actual transition into protected mode.
37
Linux Inside
{
static const struct gdt_ptr null_idt = {0, 0};
asm volatile("lidtl %0" : : "m" (null_idt));
}
which setups Interrupt descriptor table (describes interrupt handlers and etc...). For now IDT is not installed (we will see
it later), but now we just load IDT with lidtl instruction. null_idt contains address and size of IDT, but now they are just
zero. null_idt is a gdt_ptr structure, it looks:
struct gdt_ptr {
u16 len;
u32 ptr;
} __attribute__((packed));
where we can see - 16-bit length of IDT and 32-bit pointer to it (More details about IDT and interruptions we will see in the
next posts). __attribute__((packed)) means here that size of gdt_ptr minimum as required. So size of the gdt_ptr will be
6 bytes here or 48 bits. (Next we will load pointer to the gdt_ptr to the GDTR register and you can remember from the
previous post that it is 48-bits size).
For code, data and TSS (Task state segment). We will not use task state segment for now, it was added there to make Intel
VT happy as we can see in the comment line (if you're interesting you can find commit which describes it - here). Let's look
on boot_gdt . First of all we can note that it has __attribute__((aligned(16))) attribute. It means that this structure will be
aligned by 16 bytes. Let's look on simple example:
#include <stdio.h>
struct aligned {
int a;
}__attribute__((aligned(16)));
struct nonaligned {
int b;
};
int main(void)
{
struct aligned a;
struct nonaligned na;
printf("Not aligned - %zu \n", sizeof(na));
printf("Aligned - %zu \n", sizeof(a));
return 0;
}
Technically structure which contains one int field, must be 4 bytes, but here aligned structure will be 16 bytes:
Video mode initialization and transition to protected mode
38
Linux Inside
GDT_ENTRY_BOOT_CS has index - 2 here, GDT_ENTRY_BOOT_DS is GDT_ENTRY_BOOT_CS + 1 and etc... It starts from 2, because first
is a mandatory null descriptor (index - 0) and the second is not used (index - 1).
GDT_ENTRY is a macro which takes flags, base and limit and builds GDT entry. For example let's look on the code segment
in binary. Let's try to understand what every bit means. We will go through all bits from left to right:
1 - (G) granularity bit
1 - (D) if 0 16-bit segment; 1 = 32-bit segment
0 - (L) executed in 64 bit mode if 1
0 - (AVL) available for use by system software
0000 - 4 bit length 19:16 bits in the descriptor
1 - (P) segment presence in memory
00 - (DPL) - privilege level, 0 is the highest privilege
1 - (S) code or data segment, not a system segment
101 - segment type execute/read/
1 - accessed bit
You can know more about every bit in the previous post or in the Intel 64 and IA-32 Architectures Software Developers
Manuals 3A.
After this we get length of GDT with:
gdt.len = sizeof(boot_gdt)-1;
We get size of boot_gdt and subtract 1 (the last valid address in the GDT).
Next we get pointer to the GDT with:
Here we just get address of boot_gdt and add it to address of data segment shifted on 4 (remember we're in the real mode
now).
In the last we execute lgdtl instruction to load GDT into GDTR register:
39
Linux Inside
As you can read above GDT_ENTRY_BOOT_CS has index 2 and every GDT entry is 8 byte, so CS will be 2 * 8 = 16 ,
__BOOT_DS is 24 and etc... Next we set PE (Protection Enable) bit in the CR0 control register:
where 0x66 is the operand-size prefix, which allows to mix 16-bit and 32-bit code, 0xea - is the jump opcode, in_pm32 is
the segment offset and __BOOT_CS is the segment.
After this we are in the protected mode:
.code32
.section ".text32","ax"
Let's look on the first steps in the protected mode. First of all we setup data segment with:
40
Linux Inside
if you read with attention, you can remember that we saved $__BOOT_DS in the cx register. Now we fill with it all segment
registers besides cs ( cs is already __BOOT_CS ). Next we zero out all general purpose registers besides eax with:
jmpl *%eax
remember that eax contains address of the 32-bit entry (we passed it as first parameter into protected_mode_jump ). That's
all we're in the protected mode and stops before it's entry point. What is happening after we joined in the 32-bit entry point
we will see in the next part.
Conclusion
It is the end of the third part about linux kernel internals. In next part we will see first steps in the protected mode and
transition into the long mode.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
VGA
VESA BIOS Extensions
Data structure alignment
Non-maskable interrupt
A20
GCC designated inits
GCC type attributes
Previous part
41
Linux Inside
jmpl *%eax
Remind that eax register contains the address of the 32-bit entry point. We can read about this point from the linux kernel
x86 boot protocol:
And now we can make sure that it is true. Let's look on registers value in 32-bit entry point:
We can see here that cs register contains - 0x10 (as you can remember from the previous part, it is the second index in
the Global Descriptor Table), eip register is 0x100000 and base address of the all segments include code segment is zero.
So we can get physical address, it will be 0:0x100000 or just 0x100000 , as in boot protocol. Now let's start with 32-bit entry
point.
__HEAD
.code32
ENTRY(startup_32)
....
....
42
Linux Inside
....
ENDPROC(startup_32)
First of all why compressed directory? Actually bzimage is a gzipped vmlinux + header + kernel setup code . We saw the
kernel setup code in the all of previous parts. So, the main goal of the head_64.S is to prepare for entering long mode, enter
into it and decompress the kernel. We will see all of these steps besides kernel decompression in this part.
Also you can note that there are two files in the arch/x86/boot/compressed directory:
head_32.S
head_64.S
We will see only head_64.S because we are learning linux kernel for x86_64 . head_32.S even not compiled in our case.
Let's look on the arch/x86/boot/compressed/Makefile, we can see there following target:
Note on $(obj)/head_$(BITS).o . It means that compilation of the head_{32,64}.o depends on value of the $(BITS) . We can
find it in the other Makefile - arch/x86/kernel/Makefile:
ifeq ($(CONFIG_X86_32),y)
BITS := 32
...
...
else
...
...
BITS := 64
endif
__HEAD
.code32
ENTRY(startup_32)
SECTIONS
{
. = 0;
.head.text : {
_head = . ;
43
Linux Inside
HEAD_TEXT
_ehead = . ;
}
Note on . = 0; . . is a special variable of linker - location counter. Assigning a value to it, is an offset relative to the offset
of the segment. As we assign zero to it, we can read from comments:
Ok, now we know where we are, and now the best time to look inside the startup_32 function.
In the start of the startup_32 we can see the cld instruction which clears DF flag. After this, string operations like stosb
and other will increment the index registers esi or edi .
The Next we can see the check of KEEP_SEGMENTS flag from loadflags . If you remember we already saw loadflags in the
arch/x86/boot/head.S (there we checked flag CAN_USE_HEAP ). Now we need to check KEEP_SEGMENTS flag. We can find
and if KEEP_SEGMENTS is not set, we need to set ds , ss and es registers to flat segment with base 0. That we do:
remember that __BOOT_DS is 0x18 (index of data segment in the Global Descriptor Table). If KEEP_SEGMENTS is not set, we
jump to the label 1f or update segment registers with __BOOT_DS if this flag is set.
If you read previous the part, you can remember that we already updated segment registers in the
arch/x86/boot/pmjump.S, so why we need to set up it again? Actually linux kernel has also 32-bit boot protocol, so
startup_32 can be first function which will be executed right after a bootloader transfers control to the kernel.
As we checked KEEP_SEGMENTS flag and put the correct value to the segment registers, next step is calculate difference
between where we loaded and compiled to run (remember that setup.ld.S contains . = 0 at the start of the section):
Here esi register contains address of the boot_params structure. boot_params contains special field scratch with offset
0x1e4 . We are getting address of the scratch field + 4 bytes and put it to the esp register (we will use it as stack for these
calculations). After this we can see call instruction and 1f label as operand of it. What does it mean call ? It means that it
44
Linux Inside
pushes ebp value in the stack, next esp value, next function arguments and return address in the end. After this we pop
return address from the stack into ebp register ( ebp will contain return address) and subtract address of the previous label
1 .
boots_stack_end is in the .bss section, we can see definition of it in the end of head_64.S :
.bss
.balign 4
boot_heap:
.fill BOOT_HEAP_SIZE, 1, 0
boot_stack:
.fill BOOT_STACK_SIZE, 1, 0
boot_stack_end:
First of all we put address of the boot_stack_end into eax register and add to it value of the ebp (remember that ebp now
contains address where we loaded - 0x100000 ). In the end we just put eax value into esp and that's all, we have correct
stack pointer.
The next step is CPU verification. Need to check that CPU has support of long mode and SSE :
call verify_cpu
testl %eax, %eax
jnz no_longmode
It just calls verify_cpu function from the arch/x86/kernel/verify_cpu.S which contains a couple of calls of the cpuid
instruction. cpuid is instruction which is used for getting information about processor. In our case it checks long mode and
SSE support and returns 0 on success or 1 on fail in the eax register.
If eax is not zero, we jump to the no_longmode label which just stops the CPU with hlt instruction while any hardware
interrupt will not happen.
no_longmode:
1:
hlt
jmp 1b
We set stack, cheked CPU and now can move on the next step.
45
Linux Inside
The next step is calculating relocation address for decompression if need. We can see following assembly code:
#ifdef CONFIG_RELOCATABLE
movl %ebp, %ebx
movl BP_kernel_alignment(%esi), %eax
decl %eax
addl %eax, %ebx
notl %eax
andl %eax, %ebx
cmpl $LOAD_PHYSICAL_ADDR, %ebx
jge 1f
#endif
movl $LOAD_PHYSICAL_ADDR, %ebx
1:
addl $z_extract_offset, %ebx
First of all note on CONFIG_RELOCATABLE macro. This configuration option defined in the arch/x86/Kconfig and as we can read
from it's description:
In short words, this code calculates address where to move kernel for decompression put it to ebx register if the kernel is
relocatable or bzimage will decompress itself above LOAD_PHYSICAL_ADDR .
Let's look on the code. If we have CONFIG_RELOCATABLE=n in our kernel configuration file, it just puts LOAD_PHYSICAL_ADDR to
the ebx register and adds z_extract_offset to ebx . As ebx is zero for now, it will contain z_extract_offset . Now let's try
to understand these two values.
LOAD_PHYSICAL_ADDR is the macro which defined in the arch/x86/include/asm/boot.h and it looks like this:
Here we calculates aligned address where kernel is loaded ( 0x100000 or 1 megabyte in our case). PHYSICAL_ALIGN is an
alignment value to which kernel should be aligned, it ranges from 0x200000 to 0x1000000 for x86_64. With the default
values we will get 2 megabytes in the LOAD_PHYSICAL_ADDR :
After that we got alignment unit, we adds z_extract_offset (which is 0xe5c000 in my case) to the 2 megabytes. In the end
we will get 17154048 byte offset. You can find z_extract_offset in the arch/x86/boot/compressed/piggy.S . This file
generated in compile time by mkpiggy program.
Now let's try to understand the code if CONFIG_RELOCATABLE is y .
First of all we put ebp value to the ebx (remember that ebp contains address where we loaded) and kernel_alignment
field from kernel setup header to the eax register. kernel_alignment is a physical address of alignment required for the
kernel. Next we do the same as in the previous case (when kernel is not relocatable), but we just use value of the
kernel_alignment field as align unit and ebx (address where we loaded) as base address instead of CONFIG_PHYSICAL_ALIGN
46
Linux Inside
and LOAD_PHYSICAL_ADDR .
After that we calculated address, we compare it with LOAD_PHYSICAL_ADDR and add z_extract_offset to it again or put
LOAD_PHYSICAL_ADDR in the ebx if calculated address is less than we need.
After all of this calculation we will have ebp which contains address where we loaded and ebx with address where to
move kernel for decompression.
Here we put the address from ebp with gdt offset to eax register, next we put this address into ebp with offset gdt+2 and
load Global Descriptor Table with the lgdt instruction.
Let's look on Global Descriptor Table definition:
.data
gdt:
.word gdt_end - gdt
.long gdt
.word 0
.quad 0x0000000000000000 /* NULL descriptor */
.quad 0x00af9a000000ffff /* __KERNEL_CS */
.quad 0x00cf92000000ffff /* __KERNEL_DS */
.quad 0x0080890000000000 /* TS descriptor */
.quad 0x0000000000000000 /* TS continued */
It defined in the same file in the .data section. It contains 5 descriptors: null descriptor, for kernel code segment, kernel
data segment and two task descriptors. We already loaded GDT in the previous part, we're doing almost the same here, but
descriptors with CS.L = 1 and CS.D = 0 for execution in the 64 bit mode.
After we have loaded Global Descriptor Table, we must enable PAE mode with putting value of cr4 register into eax ,
setting 5 bit in it and load it again in the cr4 :
Now we finished almost with all preparations before we can move into 64-bit mode. The last step is to build page tables, but
before some information about long mode.
Long mode
Long mode is the native mode for x86_64 processors. First of all let's look on some difference between x86_64 and x86 .
It provides some features as:
47
Linux Inside
New 8 general purpose registers from r8 to r15 + all general purpose registers are 64-bit now
64-bit instruction pointer - RIP
New operating mode - Long mode
64-Bit Addresses and Operands
RIP Relative Addressing (we will see example if it in the next parts)
Long mode is an extension of legacy protected mode. It consists from two sub-modes:
64-bit mode
compatibility mode
To switch into 64-bit mode we need to do following things:
enable PAE (we already did it, see above)
build page tables and load the address of top level page table into cr3 register
enable EFER.LME
enable paging
We already enabled PAE with setting the PAE bit in the cr4 register. Now let's look on paging.
We put address which stored in ebx (remember that ebx contains the address where to relocate kernel for
decompression) with pgtable offset to the edi register. pgtable defined in the end of head_64.S and looks:
.section ".pgtable","a",@nobits
.balign 4096
pgtable:
.fill 6*4096, 1, 0
It is in the .pgtable section and it size is 24 kilobytes. After we put address to the edi , we zero out eax register and
writes zeros to the buffer with rep stosl instruction.
Now we can build top level page table - PML4 with:
48
Linux Inside
Here we get address which stored in the ebx with pgtable offset and put it to the edi . Next we put this address with offset
0x1007 to the eax register. 0x1007 is 4096 bytes (size of the PML4) + 7 (PML4 entry flags - PRESENT+RW+USER ) and puts
eax to the edi . After this manipulations edi will contain the address of the first Page Directory Pointer Entry with flags PRESENT+RW+USER .
In the next step we build 4 Page Directory entry in the Page Directory Pointer table, where first entry will be with 0x7 flags
and other with 0x8 :
We put base address of the page directory pointer table to the edi and address of the first page directory pointer entry to
the eax . Put 4 to the ecx register, it will be counter in the following loop and write the address of the first page directory
pointer table entry to the edi register.
After this edi will contain address of the first page directory pointer entry with flags 0x7 . Next we just calculates address
of following page directory pointer entries with flags 0x8 and writes their addresses to the edi .
The next step is building of 2048 page table entries by 2 megabytes:
Here we do almost the same that in the previous example, just first entry will be with flags - $0x00000183 - PRESENT + WRITE
+ MBZ and all another with 0x8 . In the end we will have 2048 pages by 2 megabytes.
Our early page table structure are done, it maps 4 gigabytes of memory and now we can put address of the high-level page
table - PML4 to the cr3 control register:
49
Linux Inside
Here we put MSR_EFER flag (which defined in the arch/x86/include/uapi/asm/msr-index.h) to the ecx register and call rdmsr
instruction which reads MSR register. After rdmsr executed, we will have result data in the edx:eax which depends on ecx
value. We check EFER_LME bit with btsl instruction and write data from eax to the MSR register with wrmsr instruction.
In next step we push address of the kernel segment code to the stack (we defined it in the GDT) and put address of the
startup_64 routine to the eax .
pushl $__KERNEL_CS
leal startup_64(%ebp), %eax
After this we push this address to the stack and enable paging with setting PG and PE bits in the cr0 register:
and call:
lret
Remember that we pushed address of the startup_64 function to the stack in the previous step, and after lret instruction,
CPU extracts address of it and jumps there.
After all of these steps we're finally in the 64-bit mode:
.code64
.org 0x200
ENTRY(startup_64)
....
....
....
That's all!
Conclusion
This is the end of the fourth part linux kernel booting process. If you have questions or suggestions, ping me in twitter
0xAX, drop me email or just create an issue.
In the next part we will see kernel decompression and many more.
Please note that English is not my first language and I am really sorry for any inconvenience. If you found any
mistakes please send me PR to linux-internals.
Links
50
Linux Inside
Protected mode
Intel 64 and IA-32 Architectures Software Developers Manual 3A
GNU linker
SSE
Paging
Model specific register
.fill instruction
Previous part
Paging on osdev.org
Paging Systems
x86 Paging Tutorial
51
Linux Inside
pushl $__KERNEL_CS
leal startup_64(%ebp), %eax
...
...
...
pushl %eax
...
...
...
lret
in the previous part, startup_64 starts to work. Since we loaded the new Global Descriptor Table and there was CPU
transition in other mode (64-bit mode in our case), we can see setup of the data segments:
.code64
.org 0x200
ENTRY(startup_64)
xorl %eax, %eax
movl %eax, %ds
movl %eax, %es
movl %eax, %ss
movl %eax, %fs
movl %eax, %gs
in the beginning of the startup_64 . All segment registers besides cs points now to the ds which is 0x18 (if you don't
understand why it is 0x18 , read the previous part).
The next step is computation of difference between where kernel was compiled and where it was loaded:
#ifdef CONFIG_RELOCATABLE
leaq startup_32(%rip), %rbp
movl BP_kernel_alignment(%rsi), %eax
decl %eax
addq %rax, %rbp
notq %rax
andq %rax, %rbp
cmpq $LOAD_PHYSICAL_ADDR, %rbp
jge 1f
#endif
movq $LOAD_PHYSICAL_ADDR, %rbp
1:
leaq z_extract_offset(%rbp), %rbx
Kernel decompression
52
Linux Inside
rbp contains decompressed kernel start address and after this code executed rbx register will contain address where to
relocate the kernel code for decompression. We already saw code like this in the startup_32 ( you can read about it in the
previous part - Calculate relocation address), but we need to do this calculation again because bootloader can use 64-bit
boot protocol and startup_32 just will not be executed in this case.
In the next step we can see setup of the stack and reset of flags register:
As you can see above rbx register contains the start address of the decompressing kernel code and we just put this
address with boot_stack_end offset to the rsp register. After this stack will be correct. You can find definition of the
boot_stack_end in the end of compressed/head_64.S file:
.bss
.balign 4
boot_heap:
.fill BOOT_HEAP_SIZE, 1, 0
boot_stack:
.fill BOOT_STACK_SIZE, 1, 0
boot_stack_end:
It located in the .bss section right before .pgtable . You can look at arch/x86/boot/compressed/vmlinux.lds.S to find it.
As we set the stack, now we can copy the compressed kernel to the address that we got above, when we calculated the
relocation address of the decompressed kernel. Let's look on this code:
pushq %rsi
leaq (_bss-8)(%rip), %rsi
leaq (_bss-8)(%rbx), %rdi
movq $_bss, %rcx
shrq $3, %rcx
std
rep movsq
cld
popq %rsi
First of all we push rsi to the stack. We need save value of rsi , because this register now stores pointer to the
boot_params real mode structure (you must remember this structure, we filled it in the start of kernel setup). In the end of
this code we'll restore pointer to the boot_params into rsi again.
The next two leaq instructions calculates effective address of the rip and rbx with _bss - 8 offset and put it to the rsi
and rdi . Why we calculate this addresses? Actually compressed kernel image located between this copying code (from
startup_32 to the current code) and the decompression code. You can verify this by looking on the linker script -
arch/x86/boot/compressed/vmlinux.lds.S:
. = 0;
.head.text : {
_head = . ;
HEAD_TEXT
_ehead = . ;
}
.rodata..compressed : {
*(.rodata..compressed)
}
.text : {
Kernel decompression
53
Linux Inside
_text = .; /* Text */
*(.text)
*(.text.*)
_etext = . ;
}
Note that .head.text section contains startup_32 . You can remember it from the previous part:
__HEAD
.code32
ENTRY(startup_32)
...
...
...
assembly
.text
relocated:
...
...
...
/*
* Do the decompression, and jump to the new kernel..
*/
...
Note that there is std instruction before data copying, it sets DF flag and it means that rsi and rdi will be decremeted
or in other words, we will crbxopy bytes in backwards.
In the end we clear DF flag with cld instruction and restore boot_params structure to the rsi .
After it we get .text section address address and jump to it:
Kernel decompression
54
Linux Inside
Here we just clear eax , put RIP relative address of the _bss to the rdi and _ebss to rcx and fill it with zeros with rep
stosq instructions.
pushq %rsi
movq $z_run_size, %r9
pushq %r9
movq %rsi, %rdi
leaq boot_heap(%rip), %rsi
leaq input_data(%rip), %rdx
movl $z_input_len, %ecx
movq %rbp, %r8
movq $z_output_len, %r9
call decompress_kernel
popq %r9
popq %rsi
Again we save rsi with pointer to boot_params structure and call decompress_kernel from the
arch/x86/boot/compressed/misc.c with seven arguments. All arguments will be passed through the registers. We finished all
preparation and now can look on the kernel decompression.
Kernel decompression
As i wrote above, decompress_kernel function is in the arch/x86/boot/compressed/misc.c source code file. This function
starts with the video/console initialization that we saw in the previous parts. This calls need if bootloaded used 32 or 64-bit
protocols. After this we store pointers to the start of the free memory and to the end of it:
free_mem_ptr = heap;
free_mem_end_ptr = heap + BOOT_HEAP_SIZE;
where heap is the second parameter of the decompress_kernel function which we got with:
boot_heap:
.fill BOOT_HEAP_SIZE, 1, 0
where BOOT_HEAP_SIZE is 0x400000 if the kernel compressed with bzip2 or 0x8000 if not.
In the next step we call choose_kernel_location function from the arch/x86/boot/compressed/aslr.c. As we can understand
from the function name it chooses memory location where to decompress the kernel image. Let's look on this function.
At the start choose_kernel_location tries to find kaslr option in the command line if CONFIG_HIBERNATION is set and nokaslr
option if this configuration option CONFIG_HIBERNATION is not set:
#ifdef CONFIG_HIBERNATION
if (!cmdline_find_option_bool("kaslr")) {
debug_putstr("KASLR disabled by default...\n");
goto out;
Kernel decompression
55
Linux Inside
}
#else
if (cmdline_find_option_bool("nokaslr")) {
debug_putstr("KASLR disabled by cmdline...\n");
goto out;
}
#endif
out:
return (unsigned char *)choice;
which just returns the output parameter which we passed to the choose_kernel_location without any changes. Let's try to
understand what is it kaslr . We can find information about it in the documentation:
kaslr/nokaslr [X86]
Enable/disable kernel and module base offset ASLR
(Address Space Layout Randomization) if built into
the kernel. When CONFIG_HIBERNATION is selected,
kASLR is disabled by default. When kASLR is enabled,
hibernation will be disabled.
It means that we can pass kaslr option to the kernel's command line and get random address for the decompressed
kernel (more about aslr you can read here).
Let's consider the case when kernel's command line contains kaslr option.
There is the call of the mem_avoid_init function from the same aslr.c source code file. This function gets the unsafe
memory regions (initrd, kernel command line and etc...). We need to know about this memory regions to not overlap them
with the kernel after decompression. For example:
Here we can see calculation of the initrd start address and size. ext_ramdisk_image is high 32-bits of the ramdisk_image
field from boot header and ext_ramdisk_size is high 32-bits of the ramdisk_size field from boot protocol:
Kernel decompression
56
Linux Inside
...
...
0C0/004 ALL ext_ramdisk_image ramdisk_image high 32bits
0C4/004 ALL ext_ramdisk_size ramdisk_size high 32bits
...
So we're taking ext_ramdisk_image and ext_ramdisk_size , shifting they left on 32 (now they will contain low 32-bits in the
high 32-bit bits) and getting start address of the initrd and size of it. After this we store these values in the mem_avoid
array which defined as:
#define MEM_AVOID_MAX 5
static struct mem_vector mem_avoid[MEM_AVOID_MAX];
struct mem_vector {
unsigned long start;
unsigned long size;
};
The next step after we collected all unsafe memory regions in the mem_avoid array will be search of the random address
which does not overlap with the unsafe regions with the find_random_addr function.
First of all we can see align of the output address in the find_random_addr function:
you can remember CONFIG_PHYSICAL_ALIGN configuration option from the previous part. This option provides the value to
which kernel should be aligned and it is 0x200000 by default. After that we got aligned output address, we go through the
memory and collect regions which are good for decompressed kernel image:
You can remember that we collected e820_entries in the second part of the Kernel booting process part 2.
First of all process_e820_entry function does some checks that e820 memory region is not non-RAM, that the start address
of the memory region is not bigger than Maximum allowed aslr offset and that memory region is not less than value of
kernel alignment:
After this, we store e820 memory region start address and the size in the mem_vector structure (we saw definition of this
structure above):
Kernel decompression
57
Linux Inside
region.start = entry->addr;
region.size = entry->size;
As we store these values, we align the region.start as we did it in the find_random_addr function and check that we didn't
get address that bigger than original memory region:
Next we get difference between the original address and aligned and check that if the last address in the memory region is
bigger than CONFIG_RANDOMIZE_BASE_MAX_OFFSET , we reduce the memory region size that end of kernel image will be less than
maximum aslr offset:
In the end we go through the all unsafe memory regions and check that this region does not overlap unsafe ares with
kernel command line, initrd and etc...:
If memory region does not overlap unsafe regions we call slots_append function with the start address of the region.
slots_append function just collects start addresses of memory regions to the slots array:
slots[slot_max++] = addr;
After process_e820_entry will be executed, we will have array of the addresses which are safe for the decompressed kernel.
Next we call slots_fetch_random function for getting random item from this array:
if (slot_max == 0)
return 0;
return slots[get_random_long() % slot_max];
where get_random_long function checks different CPU flags as X86_FEATURE_RDRAND or X86_FEATURE_TSC and chooses
Kernel decompression
58
Linux Inside
method for getting random number (it can be obtain with RDRAND instruction, Time stamp counter, programmable interval
timer and etc...). After that we got random address execution of the choose_kernel_location is finished.
Now let's back to the misc.c. After we got address for the kernel image, there need to do some checks to be sure that
gotten random address is correctly aligned and address is not wrong.
After all these checks will see the familiar message:
Decompressing Linux...
and call decompress function which will decompress the kernel. decompress function depends on what decompression
algorithm was chosen during kernel compilartion:
#ifdef CONFIG_KERNEL_GZIP
#include "../../../../lib/decompress_inflate.c"
#endif
#ifdef CONFIG_KERNEL_BZIP2
#include "../../../../lib/decompress_bunzip2.c"
#endif
#ifdef CONFIG_KERNEL_LZMA
#include "../../../../lib/decompress_unlzma.c"
#endif
#ifdef CONFIG_KERNEL_XZ
#include "../../../../lib/decompress_unxz.c"
#endif
#ifdef CONFIG_KERNEL_LZO
#include "../../../../lib/decompress_unlzo.c"
#endif
#ifdef CONFIG_KERNEL_LZ4
#include "../../../../lib/decompress_unlz4.c"
#endif
After kernel will be decompressed, the last function handle_relocations will relocate the kernel to the address that we got
from choose_kernel_location . After that kernel relocated we return from the decompress_kernel to the head_64.S . The
address of the kernel will be in the rax register and we jump on it:
jmp *%rax
Conclusion
This is the end of the fifth and the last part about linux kernel booting process. We will not see posts about kernel booting
anymore (maybe only updates in this and previous posts), but there will be many posts about other kernel internals.
Next chapter will be about kernel initialization and we will see the first steps in the linux kernel initialization code.
If you will have any questions or suggestions write me a comment or ping me in twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Kernel decompression
59
Linux Inside
Links
address space layout randomization
initrd
long mode
bzip2
RDdRand instruction
Time Stamp Counter
Programmable Interval Timers
Previous part
Kernel decompression
60
Linux Inside
Initialization
61
Linux Inside
jmp *%rax
and now we are in the kernel. There are many things to do before the kernel will start first init process. Hope we will see
all of the preparations before kernel will start in this big chapter. We will start from the kernel entry point, which is in the
arch/x86/kernel/head_64.S. We will see first preparations like early page tables initialization, switch to a new descriptor in
kernel space and many many more, before we will see the start_kernel function from the init/main.c will be called.
So let's start.
__HEAD
.code64
.globl startup_64
startup_64:
...
...
...
We can see definition of the startup_64 routine and it defined in the __HEAD section, which is just:
We can understand default virtual and physical addresses from the linker script. Note that address of the _text is location
counter which is defined as:
. = __START_KERNEL;
for x86_64 . We can find definition of the __START_KERNEL macro in the arch/x86/include/asm/page_types.h:
First steps in the kernel
62
Linux Inside
Here we can see that __START_KERNEL is the sum of the __START_KERNEL_map (which is 0xffffffff80000000 , see post about
paging) and __PHYSICAL_START . Where __PHYSICAL_START is aligned value of the CONFIG_PHYSICAL_START . So if you will not
use kASLR and will not change CONFIG_PHYSICAL_START in the configuration addresses will be following:
Physical address - 0x1000000 ;
Virtual address - 0xffffffff81000000 .
Now we know default physical and virtual addresses of the startup_64 routine, but to know actual addresses we must to
calculate it with the following code:
Here we just put the rip-relative address to the rbp register and then subtract $_text - __START_KERNEL_map from it. We
know that compiled address of the _text is 0xffffffff81000000 and __START_KERNEL_map contains 0xffffffff81000000 , so
rbp will contain physical address of the text - 0x1000000 after this calculation. We need to calculate it because kernel
can't be run on the default address, but now we know the actual physical address.
In the next step we checks that this address is aligned with:
Here we just put address to the %rax and test first bit. PMD_PAGE_MASK indicates the mask for Page middle directory (read
paging about it) and defined as:
As we can easily calculate, PMD_PAGE_SIZE is 2 megabytes. Here we use standard formula for checking alignment and if
text address is not aligned for 2 megabytes, we jump to bad_address label.
#define MAX_PHYSMEM_BITS 46
Okay, we did some early checks and now we can move on.
First steps in the kernel
63
Linux Inside
Here we need to correct early_level4_pgt and other addresses of the page table directories, because as I wrote above,
kernel can't be run at the default 0x1000000 address. rbp register contains actual address so we add to the
early_level4_pgt , level3_kernel_pgt and level2_fixmap_pgt . Let's try to understand what these labels means. First of all
NEXT_PAGE(early_level4_pgt)
.fill 511,8,0
.quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE
NEXT_PAGE(level3_kernel_pgt)
.fill L3_START_KERNEL,8,0
.quad level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE
.quad level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
NEXT_PAGE(level2_kernel_pgt)
PMDS(0, __PAGE_KERNEL_LARGE_EXEC,
KERNEL_IMAGE_SIZE/PMD_SIZE)
NEXT_PAGE(level2_fixmap_pgt)
.fill 506,8,0
.quad level1_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
.fill 5,8,0
NEXT_PAGE(level1_fixmap_pgt)
.fill 512,8,0
address and it equals 510 . After it we can see definition of two level3_kernel_pgt entries: level2_kernel_pgt and
level2_fixmap_pgt . First is simple, it is page table entry which contains pointer to the page middle directory which maps
64
Linux Inside
access rights. The second - level2_fixmap_pgt is a virtual addresses which can refer to any physical addresses even under
kernel space.
The next level2_kernel_pgt calls PDMS macro which creates 512 megabytes from the __START_KERNEL_map for kernel text
(after these 512 megabytes will be modules memory space).
Now we know Let's back to our code which is in the beginning of the section. Remember that rbp contains actual physical
address of the _text section. We just add this address to the base address of the page tables, that they'll have correct
addresses:
At the first line we add rbp to the early_level4_pgt , at the second line we add rbp to the level2_kernel_pgt , at the third
line we add rbp to the level2_fixmap_pgt and add rbp to the level1_fixmap_pgt .
After all of this we will have:
As we corrected base addresses of the page tables, we can start to build it.
After this we store physical address of the _text in the rax and get the index of the page global directory entry which
stores _text address, by shifting _text address on the PGDIR_SHIFT :
where PGDIR_SHIFT is 39 . PGDIR_SHFT indicates the mask for page global directory bits in a virtual address. There are
macro for all types of page directories:
#define PGDIR_SHIFT 39
#define PUD_SHIFT 30
#define PMD_SHIFT 21
65
Linux Inside
After this we put the address of the first level3_kernel_pgt to the rdx with the _KERNPG_TABLE access rights (see above)
and fill the early_level4_pgt with the 2 level3_kernel_pgt entries.
After this we add 4096 (size of the early_level4_pgt ) to the rdx (it now contains the address of the first entry of the
level3_kernel_pgt ) and put rdi (it now contains physical address of the _text ) to the rax . And after this we write
In the next step we write addresses of the page middle directory entries to the level2_kernel_pgt and the last step is
correcting of the kernel text+data virtual addresses:
Here we put the address of the level2_kernel_pgt to the rdi and address of the page table entry to the r8 register. Next
we check the present bit in the level2_kernel_pgt and if it is zero we're moving to the next page by adding 8 bytes to rdi
which contaitns address of the level2_kernel_pgt . After this we compare it with r8 (contains address of the page table
entry) and go back to label 1 or move forward.
In the next step we correct phys_base physical address with rbp (contains physical address of the _text ), put physical
address of the early_level4_pgt and jump to label 1 :
where phys_base mathes the first entry of the level2_kernel_pgt which is 512 MB kernel mapping.
Last preparations
After that we jumped to the label 1 we enable PAE , PGE (Paging Global Extension) and put the physical address of the
phys_base (see above) to the rax register and fill cr3 register with it:
1:
movl $(X86_CR4_PAE | X86_CR4_PGE), %ecx
movq %rcx, %cr4
addq phys_base(%rip), %rax
movq %rax, %cr3
66
Linux Inside
We put 0x80000001 value to the eax and execute cpuid instruction for getting extended processor info and feature bits.
The result will be in the edx register which we put to the edi .
Now we put 0xc0000080 or MSR_EFER to the ecx and call rdmsr instruction for the reading model specific register.
The result will be in the edx:eax . General view of the EFER is following:
63 32
-------------------------------------------------------------------------------| |
| Reserved MBZ |
| |
-------------------------------------------------------------------------------31 16 15 14 13 12 11 10 9 8 7 1 0
-------------------------------------------------------------------------------| | T | | | | | | | | | |
| Reserved MBZ | C | FFXSR | LMSLE |SVME|NXE|LMA|MBZ|LME|RAZ|SCE|
| | E | | | | | | | | | |
--------------------------------------------------------------------------------
We will not see all fields in details here, but we will learn about this and other MSRs in the special part about. As we read
EFER to the edx:eax , we checks _EFER_SCE or zero bit which is System Call Extensions with btsl instruction and set it to
one. By the setting SCE bit we enable SYSCALL and SYSRET instructions. In the next step we check 20th bit in the edi ,
remember that this register stores result of the cpuid (see above). If 20 bit is set ( NX bit) we just write EFER_SCE to the
model specific register.
If NX bit is supported we enable _EFER_NX and write it too, with the wrmsr instruction.
In the next step we need to update Global Descriptor table with lgdt instruction:
lgdt early_gdt_descr(%rip)
early_gdt_descr:
.word GDT_ENTRIES*8-1
early_gdt_descr_base:
.quad INIT_PER_CPU_VAR(gdt_page)
67
Linux Inside
We need to reload Global Descriptor Table because now kernel works in the userspace addresses, but soon kernel will
work in it's own space. Now let's look on early_gdt_descr definition. Global Descriptor Table contains 32 entries:
#define GDT_ENTRIES 32
for kernel code, data, thread local storage segments and etc... it's simple. Now let's look on the early_gdt_descr_base . First
of gdt_page defined as:
struct gdt_page {
struct desc_struct gdt[GDT_ENTRIES];
} __attribute__((aligned(PAGE_SIZE)));
in the arch/x86/include/asm/desc.h. It contains one field gdt which is array of the desc_struct structures which defined as:
struct desc_struct {
union {
struct {
unsigned int a;
unsigned int b;
};
struct {
u16 limit0;
u16 base0;
unsigned base1: 8, type: 4, s: 1, dpl: 2, p: 1;
unsigned limit: 4, avl: 1, l: 1, d: 1, g: 1, base2: 8;
};
};
} __attribute__((packed));
and presents familiar to us GDT descriptor. Also we can note that gdt_page structure aligned to PAGE_SIZE which is 4096
bytes. It means that gdt will occupy one page. Now let's try to understand what is it INIT_PER_CPU_VAR . INIT_PER_CPU_VAR is
a macro which defined in the arch/x86/include/asm/percpu.h and just concats init_per_cpu__ with the given parameter:
As we got init_per_cpu__gdt_page in INIT_PER_CPU_VAR and INIT_PER_CPU macro from linker script will be expanded we will
get offset from the __per_cpu_load . After this calculations, we will have correct base address of the new GDT.
Generally per-CPU variables is a 2.6 kernel feature. You can understand what is it from it's name. When we create per-CPU
variable, each CPU will have will have it's own copy of this variable. Here we creating gdt_page per-CPU variable. There
are many advantages for variables of this type, like there are no locks, because each CPU works with it's own copy of
variable and etc... So every core on multiprocessor will have it's own GDT table and every entry in the table will represent a
memory segment which can be accessed from the thread which ran on the core. You can read in details about per-CPU
variables in the Theory/per-cpu post.
As we loaded new Global Descriptor Table, we reload segments as we did it every time:
xorl %eax,%eax
68
Linux Inside
movl %eax,%ds
movl %eax,%ss
movl %eax,%es
movl %eax,%fs
movl %eax,%gs
After all of these steps we set up gs register that it post to the irqstack (we will see information about it in the next parts):
movl $MSR_GS_BASE,%ecx
movl initial_gs(%rip),%eax
movl initial_gs+4(%rip),%edx
wrmsr
We need to put MSR_GS_BASE to the ecx register and load data from the eax and edx (which are point to the initial_gs )
with wrmsr instruction. We don't use cs , fs , ds and ss segment registers for addressation in the 64-bit mode, but fs
and gs registers can be used. fs and gs have a hidden part (as we saw it in the real mode for cs ) and this part contains
descriptor which mapped to Model specific registers. So we can see above 0xc0000101 is a gs.base MSR address.
In the next step we put the address of the real mode bootparam structure to the rdi (remember rsi holds pointer to this
structure from the start) and jump to the C code with:
movq initial_code(%rip),%rax
pushq $0
pushq $__KERNEL_CS
pushq %rax
lretq
Here we put the address of the initial_code to the rax and push fake address, __KERNEL_CS and the address of the
initial_code to the stack. After this we can see lretq instruction which means that after it return address will be extracted
from stack (now there is address of the initial_code ) and jump there. initial_code defined in the same source code file
and looks:
__REFDATA
.balign 8
GLOBAL(initial_code)
.quad x86_64_start_kernel
...
...
...
As we can see initial_code contains address of the x86_64_start_kernel , which defined in the arch/x86/kerne/head64.c
and looks like this:
It has one argument is a real_mode_data (remember that we passed address of the real mode data to the rdi register
First steps in the kernel
69
Linux Inside
previously).
This is first C code in the kernel!
Next to start_kernel
We need to see last preparations before we can see "kernel entry point" - start_kernel function from the init/main.c.
First of all we can see some checks in the x86_64_start_kernel function:
There are checks for different things like virtual addresses of modules space is not fewer than base address of the kernel
text - __STAT_KERNEL_map , that kernel text with modules is not less than image of the kernel and etc... BUILD_BUG_ON is a
macro which looks as:
Let's try to understand this trick works. Let's take for example first condition: MODULES_VADDR < __START_KERNEL_map .
!!conditions is the same that condition != 0 . So it means if MODULES_VADDR < __START_KERNEL_map is true, we will get 1 in
the !!(condition) or zero if not. After 2*!!(condition) we will get or 2 or 0 . In the end of calculations we can get two
different behaviors:
We will have compilation error, because try to get size of the char array with negative index (as can be in our case,
because MODULES_VADDR can't be less than __START_KERNEL_map will be in our case);
No compilation errors.
That's all. So interesting C trick for getting compile error which depends on some constants.
In the next step we can see call of the cr4_init_shadow function which stores shadow copy of the cr4 per cpu. Context
switches can change bits in the cr4 so we need to store cr4 for each CPU. And after this we can see call of the
reset_early_page_tables function where we resets all page global directory entries and write new pointer to the PGT in
cr3 :
soon we will build new page tables. Here we can see that we go through all Page Global Directory Entries ( PTRS_PER_PGD is
512 ) in the loop and make it zero. After this we set next_early_pgt to zero (we will see details about it in the next post) and
write physical address of the early_level4_pgt to the cr3 . __pa_nodebug is a macro which will be expanded to:
70
Linux Inside
After this we clear _bss from the __bss_stop to __bss_start and the next step will be setup of the early IDT handlers, but
it's big theme so we will see it in the next part.
Conclusion
This is the end of the first part about linux kernel initialization.
If you have questions or suggestions, feel free to ping me in twitter 0xAX, drop me email or just create issue.
In the next part we will see initialization of the early interruption handlers, kernel space memory mapping and many many
more.
Please note that English is not my first language and I am really sorry for any inconvenience. If you found any
mistakes please send me PR to linux-internals.
Links
Model Specific Register
Paging
Previous part - Kernel decompression
NX
ASLR
71
Linux Inside
from the arch/x86/kernel/head64.c source code file. But before we started to sort out this code, we need to know about
interrupts and handlers.
Some theory
Interrupt is an event caused by software or hardware to the CPU. On interrupt, CPU stops the current task and transfer
control to the interrupt handler, which handles interruption and transfer control back to the previously stopped task. We can
split interrupts on three types:
Software interrupts - when a software signals CPU that it needs kernel attention. These interrupts generally used for
system calls;
Hardware interrupts - when a hardware, for example button pressed on a keyboard;
Exceptions - interrupts generated by CPU, when the CPU detects error, for example division by zero or accessing a
memory page which is not in RAM.
Every interrupt and exception is assigned an unique number which called - vector number . Vector number can be any
number from 0 to 255 . There is common practice to use first 32 vector numbers for exceptions, and vector numbers from
31 to 255 are used for user-defined interrupts. We can see it in the code above - NUM_EXCEPTION_VECTORS , which defined
as:
#define NUM_EXCEPTION_VECTORS 32
CPU uses vector number as an index in the Interrupt Descriptor Table (we will see description of it soon). CPU catch
interrupts from the APIC or through it's pins. Following table shows 0-31 exceptions:
72
Linux Inside
To react on interrupt CPU uses special structure - Interrupt Descriptor Table or IDT. IDT is an array of 8-byte descriptors like
Global Descriptor Table, but IDT entries are called gates . CPU multiplies vector number on 8 to find index of the IDT entry.
But in 64-bit mode IDT is an array of 16-byte descriptors and CPU multiplies vector number on 16 to find index of the entry
in the IDT. We remember from the previous part that CPU uses special GDTR register to locate Global Descriptor Table, so
CPU uses special register IDTR for Interrupt Descriptor Table and lidt instruuction for loading base address of the table
into this register.
64-bit mode IDT entry has following structure:
127 96
-------------------------------------------------------------------------------| |
| Reserved |
| |
-------------------------------------------------------------------------------95 64
-------------------------------------------------------------------------------| |
| Offset 63..32 |
| |
-------------------------------------------------------------------------------63 48 47 46 44 42 39 34 32
-------------------------------------------------------------------------------| | | D | | | | | | |
| Offset 31..16 | P | P | 0 |Type |0 0 0 | 0 | 0 | IST |
| | | L | | | | | | |
-------------------------------------------------------------------------------31 15 16 0
-------------------------------------------------------------------------------| | |
| Segment Selector | Offset 15..0 |
| | |
--------------------------------------------------------------------------------
Where:
73
Linux Inside
We're filling only first 32 IDT entries because all of the early setup runs with interrupts disabled, so there is no need to set
up early exception handlers for vectors greater than 32. early_idt_handlers contains generic idt handlers and we can find it
in the arch/x86/kernel/head_64.S, we will look it soon.
Now let's look on set_intr_gate implementation:
74
Linux Inside
First of all it checks with that passed interrupt number is not greater than 255 with BUG_ON macro. We need to do this
check because we can have only 256 interrupts. After this it calls _set_gate which writes address of an interrupt gate to the
IDT :
At the start of _set_gate function we can see call of the pack_gate function which fills gate_desc structure with the given
values:
static inline void pack_gate(gate_desc *gate, unsigned type, unsigned long func,
unsigned dpl, unsigned ist, unsigned seg)
{
gate->offset_low = PTR_LOW(func);
gate->segment = __KERNEL_CS;
gate->ist = ist;
gate->p = 1;
gate->dpl = dpl;
gate->zero0 = 0;
gate->zero1 = 0;
gate->type = type;
gate->offset_middle = PTR_MIDDLE(func);
gate->offset_high = PTR_HIGH(func);
}
As mentioned above we fill gate descriptor in this function. We fill three parts of the address of the interrupt handler with the
address which we got in the main loop (address of the interrupt handler entry point). We are using three following macro to
split address on three parts:
With the first PTR_LOW macro we get the first 2 bytes of the address, with the second PTR_MIDDLE we get the second 2 bytes
of the address and with the third PTR_HIGH macro we get the last 4 bytes of the address. Next we setup the segment
selector for interrupt handler, it will be our kernel code segment - __KERNEL_CS . In the next step we fill Interrupt Stack
Table and Descriptor Privilege Level (highest privilege level) with zeros. And we set GAT_INTERRUPT type in the end.
Now we have filled IDT entry and we can call native_write_idt_entry function which just copies filled IDT entry to the IDT :
static inline void native_write_idt_entry(gate_desc *idt, int entry, const gate_desc *gate)
{
memcpy(&idt[entry], gate, sizeof(*gate));
}
75
Linux Inside
After that main loop will finished, we will have filled idt_table array of gate_desc structures and we can load IDT with:
You can note that there are calls of the _trace_* functions in the _set_gate and other functions. These functions fills IDT
gates in the same manner that _set_gate but with one difference. These functions use trace_idt_table Interrupt
Descriptor Table instead of idt_table for tracepoints (we will cover this theme in the another part).
Okay, now we have filled and loaded Interrupt Descriptor Table, we know how the CPU acts during interrupt. So now time to
deal with interrupts handlers.
.globl early_idt_handlers
early_idt_handlers:
i = 0
.rept NUM_EXCEPTION_VECTORS
.if (EXCEPTION_ERRCODE_MASK >> i) & 1
ASM_NOP2
.else
pushq $0
.endif
pushq $i
jmp early_idt_handler
i = i + 1
.endr
We can see here, interrupt handlers generation for the first 32 exceptions. We check here, if exception has error code then
we do nothing, if exception does not return error code, we push zero to the stack. We do it for that would stack was uniform.
After that we push exception number on the stack and jump on the early_idt_handler which is generic interrupt handler for
now. As i wrote above, CPU pushes flag register, CS and RIP on the stack. So before early_idt_handler will be executed,
stack will contain following data:
|--------------------|
| %rflags |
| %cs |
| %rip |
| rsp --> error code |
|--------------------|
Now let's look on the early_idt_handler implementation. It locates in the same arch/x86/kernel/head_64.S. First of all we
76
Linux Inside
can see check for NMI, we no need to handle it, so just ignore they in the early_idt_handler :
cmpl $2,(%rsp)
je is_nmi
where is_nmi :
is_nmi:
addq $16,%rsp
INTERRUPT_RETURN
we drop error code and vector number from the stack and call INTERRUPT_RETURN which is just iretq . As we checked the
vector number and it is not NMI , we check early_recursion_flag to prevent recursion in the early_idt_handler and if it's
correct we save general registers on the stack:
pushq %rax
pushq %rcx
pushq %rdx
pushq %rsi
pushq %rdi
pushq %r8
pushq %r9
pushq %r10
pushq %r11
we need to do it to prevent wrong values in it when we return from the interrupt handler. After this we check segment
selector in the stack:
cmpl $__KERNEL_CS,96(%rsp)
jne 11f
it must be equal to the kernel code segment and if it is not we jump on label 11 which prints PANIC message and makes
stack dump.
After code segment was checked, we check the vector number, and if it is #PF , we put value from the cr2 to the rdi
register and call early_make_pgtable (well see it soon):
cmpl $14,72(%rsp)
jnz 10f
GET_CR2_INTO(%rdi)
call early_make_pgtable
andl %eax,%eax
jz 20f
If vector number is not #PF , we restore general purpose registers from the stack:
popq %r11
popq %r10
popq %r9
popq %r8
popq %rdi
popq %rsi
popq %rdx
popq %rcx
popq %rax
77
Linux Inside
to add ability to load kernel above 4G and make access to boot_params structure above the 4G.
You can find implementation of the early_make_pgtable in the arch/x86/kernel/head64.c and takes one parameter - address
from the cr2 register, which caused Page Fault. Let's look on it:
It starts from the definition of some variables which have *val_t types. All of these types are just:
Also we will operate with the *_t (not val) types, for example pgd_t and etc... All of these types defined in the
arch/x86/include/asm/pgtable_types.h and represent structures like this:
For example,
Here early_level4_pgt presents early top-level page table directory which consists of an array of pgd_t types and pgd
points to low-level page entries.
After we made the check that we have no invalid address, we're getting the address of the Page Global Directory entry
which contains #PF address and put it's value to the pgd variable:
pgd_p = &early_level4_pgt[pgd_index(address)].pgd;
pgd = *pgd_p;
In the next step we check pgd , if it contains correct page global directory entry we put physical address of the page global
directory entry and put it to the pud_p with:
Early interrupts handler
78
Linux Inside
or
0b1111111111111111111111111111111111111111111111
we create new page upper directory pointer which points to the current dynamic page table and writes it's physical address
with the _KERPG_TABLE access rights to the page global directory:
pud_p += pud_index(address);
pud = *pud_p;
In the next step we do the same actions as we did before, but with the page middle directory. In the end we fix address of
the page middle directory which contains maps kernel text+data virtual addresses:
After page fault handler finished it's work and as result our early_level4_pgt contains entries which point to the valid
addresses.
Conclusion
Early interrupts handler
79
Linux Inside
This is the end of the second part about linux kernel internals. If you have questions or suggestions, ping me in twitter
0xAX, drop me email or just create issue. In the next part we will see all steps before kernel entry point - start_kernel
function.
Please note that English is not my first language and I am really sorry for any inconvenience. If you found any
mistakes please send me PR to linux-internals.
Links
GNU assembly .rept
APIC
NMI
Previous part
80
Linux Inside
boot_params again
In the previous part we stopped at setting Interrupt Descriptor Table and loading it in the IDTR register. At the next step
after this we can see a call of the copy_bootdata function:
copy_bootdata(__va(real_mode_data));
This function takes one argument - virtual address of the real_mode_data . Remember that we passed the address of the
boot_params structure from arch/x86/include/uapi/asm/bootparam.h to the x86_64_start_kernel function as first argument in
arch/x86/kernel/head_64.S:
where PAGE_OFFSET is __PAGE_OFFSET which is 0xffff880000000000 and the base virtual address of the direct mapping of all
physical memory. So we're getting virtual address of the boot_params structure and pass it to the copy_bootdata function,
where we copy real_mod_data to the boot_params which is declared in the arch/x86/kernel/setup.h
81
Linux Inside
First of all, note that this function is declared with __init prefix. It means that this function will be used only during the
initialization and used memory will be freed.
We can see declaration of two variables for the kernel command line and copying real_mode_data to the boot_params with
the memcpy function. The next call of the sanitize_boot_params function which fills some fields of the boot_params structure
like ext_ramdisk_image and etc... if bootloaders which fail to initialize unknown fields in boot_params to zero. After this we're
getting address of the command line with the call of the get_cmd_line_ptr function:
which gets the 64-bit address of the command line from the kernel boot header and returns it. In the last step we check that
we got cmd_line_pty , getting its virtual address and copy it to the boot_command_line which is just an array of bytes:
After this we will have copied kernel command line and boot_params structure. In the next step we can see call of the
load_ucode_bsp function which loads processor microcode, but we will not see it here.
After microcode was loaded we can see the check of the console_loglevel and the early_printk function which prints
Kernel Alive string. But you'll never see this output because early_printk is not initilized yet. It is a minor bug in the
kernel and i sent the patch - commit and you will see it in the mainline soon. So you can skip this code.
clear_page(init_level4_pgt);
function and pass init_level4_pgt which defined also in the arch/x86/kernel/head_64.S and looks:
NEXT_PAGE(init_level4_pgt)
.quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
.org init_level4_pgt + L4_PAGE_OFFSET*8, 0
.quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
.org init_level4_pgt + L4_START_KERNEL*8, 0
.quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE
which maps first 2 gigabytes and 512 megabytes for the kernel code, data and bss. clear_page function defined in the
arch/x86/lib/clear_page_64.S let look on this function:
ENTRY(clear_page)
CFI_STARTPROC
xorl %eax,%eax
movl $4096/64,%ecx
82
Linux Inside
.p2align 4
.Lloop:
decl %ecx
#define PUT(x) movq %rax,x*8(%rdi)
movq %rax,(%rdi)
PUT(1)
PUT(2)
PUT(3)
PUT(4)
PUT(5)
PUT(6)
PUT(7)
leaq 64(%rdi),%rdi
jnz .Lloop
nop
ret
CFI_ENDPROC
.Lclear_page_end:
ENDPROC(clear_page)
As you can understart from the function name it clears or fills with zeros page tables. First of all note that this function starts
with the CFI_STARTPROC and CFI_ENDPROC which are expands to GNU assembly directives:
and used for debugging. After CFI_STARTPROC macro we zero out eax register and put 64 to the ecx (it will be counter).
Next we can see loop which starts with the .Lloop label and it starts from the ecx decrement. After it we put zero from the
rax register to the rdi which contains the base address of the init_level4_pgt now and do the same procedure seven
times but every time move rdi offset on 8. After this we will have first 64 bytes of the init_level4_pgt filled with zeros. In
the next step we put the address of the init_level4_pgt with 64-bytes offset to the rdi again and repeat all operations
which ecx is not zero. In the end we will have init_level4_pgt filled with zeros.
As we have init_level4_pgt filled with zeros, we set the last init_level4_pgt entry to kernel high mapping with the:
init_level4_pgt[511] = early_level4_pgt[511];
Remember that we dropped all early_level4_pgt entries in the reset_early_page_table function and kept only kernel high
mapping there.
The last step in the x86_64_start_kernel function is the call of the:
x86_64_start_reservations(real_mode_data);
function with the real_mode_data as argument. The x86_64_start_reservations function defined in the same source code
file as the x86_64_start_kernel function and looks:
83
Linux Inside
You can see that it is the last function before we are in the kernel entry point - start_kernel function. Let's look what it does
and how it works.
if (!boot_params.hdr.version)
copy_bootdata(__va(real_mode_data));
and if it is not we call again copy_bootdata function with the virtual address of the real_mode_data (read about about it's
implementation).
In the next step we can see the call of the reserve_ebda_region function which defined in the arch/x86/kernel/head.c. This
function reserves memory block for th EBDA or Extended BIOS Data Area. The Extended BIOS Data Area located in the top
of conventional memory and contains data about ports, disk parameters and etc...
Let's look on the reserve_ebda_region function. It starts from the checking is paravirtualization enabled or not:
if (paravirt_enabled())
return;
we exit from the reserve_ebda_region function if paravirtualization is enabled because if it enabled the extended bios data
area is absent. In the next step we need to get the end of the low memory:
We're getting the virtual address of the BIOS low memory in kilobytes and convert it to bytes with shifting it on 10 (multiply
on 1024 in other words). After this we need to get the address of the extended BIOS data are with the:
ebda_addr = get_bios_ebda();
Let's try to understand how it works. Here we can see that we converting physical address 0x40E to the virtual, where
0x0040:0x000e is the segment which contains base address of the extended BIOS data area. Don't worry that we are using
phys_to_virt function for converting a physical address to virtual address. You can note that previously we have used
__va macro for the same point, but phys_to_virt is the same:
84
Linux Inside
only with one difference: we pass argument with the phys_addr_t which depends on CONFIG_PHYS_ADDR_T_64BIT :
#ifdef CONFIG_PHYS_ADDR_T_64BIT
typedef u64 phys_addr_t;
#else
typedef u32 phys_addr_t;
#endif
This configuration option is enabled by CONFIG_PHYS_ADDR_T_64BIT . After that we got virtual address of the segment which
stores the base address of the extended BIOS data area, we shift it on 4 and return. After this ebda_addr variables contains
the base address of the extended BIOS data area.
In the next step we check that address of the extended BIOS data area and low memory is not less than INSANE_CUTOFF
macro
which is:
or 128 kilobytes. In the last step we get lower part in the low memory and extended bios data area and call
memblock_reserve function which will reserve memory region for extended bios data between low memory and one
megabyte mark:
85
Linux Inside
struct memblock_type {
unsigned long cnt;
unsigned long max;
phys_addr_t total_size;
struct memblock_region *regions;
};
As we need to reserve memory block for extended bios data area, the type of the current memory region is reserved where
memblock structure is:
struct memblock {
bool bottom_up;
phys_addr_t current_limit;
struct memblock_type memory;
struct memblock_type reserved;
#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
struct memblock_type physmem;
#endif
};
and describes generic memory block. You can see that we initialize _rgn by assigning it to the address of the
memblock.reserved . memblock is the global variable which looks:
We will not dive into detail of this varaible, but we will see all details about it in the parts about memory manager. Just note
that memblock variable defined with the __initdata_memblock which is:
86
Linux Inside
From this we can conclude that all memory blocks will be in the .meminit.data section. After we defined _rgn we print
information about it with memblock_dbg macros. You can enable it by passing memblock=debug to the kernel command line.
After debugging lines were printed next is the call of the following function:
which adds new memory block region into the .meminit.data section. As we do not initlieze _rgn but it just contains
&memblock.reserved , we just fill passed _rgn with the base address of the extended BIOS data area region, size of this
if (type->regions[0].size == 0) {
WARN_ON(type->cnt != 1 || type->total_size);
type->regions[0].base = base;
type->regions[0].size = size;
type->regions[0].flags = flags;
memblock_set_region_node(&type->regions[0], nid);
type->total_size = size;
return 0;
}
After we filled our region we can see the call of the memblock_set_region_node function with two parameters:
address of the filled memory region;
NUMA node id.
where our regions represented by the memblock_region structure:
struct memblock_region {
phys_addr_t base;
phys_addr_t size;
unsigned long flags;
#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
int nid;
#endif
};
#ifdef CONFIG_NODES_SHIFT
#define NODES_SHIFT CONFIG_NODES_SHIFT
#else
#define NODES_SHIFT 0
#endif
87
Linux Inside
memblick_set_region_node function just fills nid field from memblock_region with the given value:
After this we will have first reserved memblock for the extended bios data area in the .meminit.data section.
reserve_ebda_region function finished its work on this step and we can go back to the arch/x86/kernel/head64.c.
We finished all preparations before the kernel entry point! The last step in the x86_64_start_reservations function is the call
of the:
start_kernel()
Conclusion
It is the end of the third part about linux kernel internals. In next part we will see the first initialization steps in the kernel
entry point - start_kernel function. It will be the first step before we will see launch of the first init process.
If you have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
BIOS data area
What is in the extended BIOS data area on a PC?
Previous part
88
Linux Inside
After the initialization process will be finished, the kernel will release these sections with a call to the free_initmem function.
Note also that __init is defined with two attributes: __cold and notrace . The purpose of the first cold attribute is to mark
that the function is rarely used and the compiler must optimize this function for size. The second notrace is defined as:
where no_instrument_function says to the compiler not to generate profiling function calls.
In the definition of the start_kernel function, you can also see the __visible attribute which expands to the:
where externally_visible tells to the compiler that something uses this function or variable, to prevent marking this
function/variable as unusable . You can find the definition of this and other macro attributes in include/linux/init.h.
89
Linux Inside
At the beginning of the start_kernel you can see the definition of these two variables:
char *command_line;
char *after_dashes;
The first represents a pointer to the kernel command line and the second will contain the result of the parse_args function
which parses an input string with parameters in the form name=value , looking for specific keywords and invoking the right
handlers. We will not go into the details related with these two variables at this time, but will see it in the next parts. In the
next step we can see a call to the:
lockdep_init();
function. lockdep_init initializes lock validator. Its implementation is pretty simple, it just initializes two list_head hashes
and sets the lockdep_initialized global variable to 1 . Lock validator detects circular lock dependencies and is called
when any spinlock or mutex is acquired.
The next function is set_task_stack_end_magic which takes address of the init_task and sets STACK_END_MAGIC
( 0x57AC6E9D ) as canary for it. init_task represents the initial task structure:
where task_struct stores all the information about a process. I will not explain this structure in this book because it's very
big. You can find its definition in include/linux/sched.h. At this moment task_struct contains more than 100 fields!
Although you will not see the explanation of the task_struct in this book, we will use it very often since it is the
fundamental structure which describes the process in the Linux kernel. I will describe the meaning of the fields of this
structure as we meet them in practice.
You can see the definition of the init_task and it initialized by the INIT_TASK macro. This macro is from
include/linux/init_task.h and it just fills the init_task with the values for the first process. For example it sets:
init process state to zero or runnable . A runnable process is one which is waiting only for a CPU to run on;
init process flags - PF_KTHREAD which means - kernel thread;
a list of runnable task;
process address space;
init process stack to the &init_thread_info which is init_thread_union.thread_info and initthread_union has type thread_union which contains thread_info and process stack:
union thread_union {
struct thread_info thread_info;
unsigned long stack[THREAD_SIZE/sizeof(long)];
};
Every process has its own stack and it is 16 killobytes or 4 page frames. in x86_64 . We can note that it is defined as array
of unsigned long . The next field of the thread_union is - thread_info defined as:
struct thread_info {
struct task_struct *task;
struct exec_domain *exec_domain;
__u32 flags;
__u32 status;
__u32 cpu;
int saved_preempt_count;
90
Linux Inside
mm_segment_t addr_limit;
struct restart_block restart_block;
void __user *sysenter_return;
unsigned int sig_on_uaccess_error:1;
unsigned int uaccess_err:1;
};
and occupies 52 bytes. The thread_info structure contains architecture-specific information on the thread. We know that
on x86_64 the stack grows down and thread_union.thread_info is stored at the bottom of the stack in our case. So the
process stack is 16 killobytes and thread_info is at the bottom. The remaining thread_size will be 16 killobytes - 62 bytes
= 16332 bytes . Note that thread_unioun represented as the union and not structure, it means that thread_info and stack
+-----------------------+
| |
| |
| stack |
| |
|_______________________|
| | |
| | |
| | |
|______________________| +--------------------+
| | | |
| thread_info |<----------->| task_struct |
| | | |
+-----------------------+ +--------------------+
http://www.quora.com/In-Linux-kernel-Why-thread_info-structure-and-the-kernel-stack-of-a-process-binds-in-unionconstruct
So the INIT_TASK macro fills these task_struct's fields and many many more. As I already wrote about, I will not describe
all the fields and values in the INIT_TASK macro but we will see them soon.
Now let's go back to the set_task_stack_end_magic function. This function defined in the kernel/fork.c and sets a canary to
the init process stack to prevent stack overflow.
Its implementation is simple. set_task_stack_end_magic gets the end of the stack for the given task_struct with the
end_of_stack function. The end of a process stack depends on the CONFIG_STACK_GROWSUP configuration option. As we learn
in x86_64 architecture, the stack grows down. So the end of the process stack will be:
where task_thread_info just returns the stack which we filled with the INIT_TASK macro:
91
Linux Inside
As we got the end of the init process stack, we write STACK_END_MAGIC there. After canary is set, we can check it like this:
if (*end_of_stack(task) != STACK_END_MAGIC) {
//
// handle stack overflow here
//
}
The next function after the set_task_stack_end_magic is smp_setup_processor_id . This function has an empty body for
x86_64 :
as it not implemented for all architectures, but some such as s390 and arm64.
The next function in start_kernel is debug_objects_early_init . Implementation of this function is almost the same as
lockdep_init , but fills hashes for object debugging. As I wrote about, we will not see the explanation of this and other
otherwise it generates random numbers based on random pool and the TSC:
get_random_bytes(&canary, sizeof(canary));
tsc = __native_read_tsc();
canary += tsc + (tsc << 32UL);
After we got a random number, we fill the stack_canary field of task_struct with it:
current->stack_canary = canary;
and write this value to the top of the IRQ stack with the:
Again, we will not dive into details here, we will cover it in the part about IRQs. As canary is set, we disable local and early
boot IRQs and register the bootstrap CPU in the CPU maps. We disable local IRQs (interrupts for current CPU) with the
local_irq_disable macro which expands to the call of the arch_local_irq_disable function from include/linux/percpu-
defs.h:
Where native_irq_enable is cli instruction for x86_64 . As interrupts are disabled we can register the current CPU with
the given ID in the CPU bitmap.
92
Linux Inside
For now it is just zero. If the CONFIG_DEBUG_PREEMPT configuration option is disabled, smp_processor_id just expands to the
call of raw_smp_processor_id which expands to the:
this_cpu_read as many other function like this ( this_cpu_write , this_cpu_add and etc...) defined in the
include/linux/percpu-defs.h and presents this_cpu operation. These operations provide a way of optimizing access to the
per-cpu variables which are associated with the current processor. In our case it is this_cpu_read :
__pcpu_size_call_return(this_cpu_read_, pcp)
Remember that we have passed cpu_number as pcp to the this_cpu_read from the raw_smp_processor_id . Now let's look at
the __pcpu_size_call_return implementation:
Yes, it looks a little strange but it's easy. First of all we can see the definition of the pscr_ret__ variable with the int type.
Why int? Ok, variable is common_cpu and it was declared as per-cpu int variable:
DECLARE_PER_CPU_READ_MOSTLY(int, cpu_number);
In the next step we call __verify_pcpu_ptr with the address of cpu_number . __veryf_pcpu_ptr used to verify that the given
parameter is a per-cpu pointer. After that we set pscr_ret__ value which depends on the size of the variable. Our
common_cpu variable is int , so it 4 bytes in size. It means that we will get this_cpu_read_4(common_cpu) in pscr_ret__ . In
which calls percpu_from_op and pass mov instruction and per-cpu variable there. percpu_from_op will expand to the inline
assembly call:
Kernel entry point
93
Linux Inside
Let's try to understand how it works and what it does. The gs segment register contains the base of per-cpu area. Here we
just copy common_cpu which is in memory to the pfo_ret__ with the movl instruction. Or with another words:
this_cpu_read(common_cpu)
As we didn't setup per-cpu area, we have only one - for the current running CPU, we will get zero as a result of the
smp_processor_id .
As we got the current processor id, boot_cpu_init sets the given CPU online, active, present and possible with the:
set_cpu_online(cpu, true);
set_cpu_active(cpu, true);
set_cpu_present(cpu, true);
set_cpu_possible(cpu, true);
All of these functions use the concept - cpumask . cpu_possible is a set of CPU ID's which can be plugged in at any time
during the life of that system boot. cpu_present represents which CPUs are currently plugged in. cpu_online represents
subset of the cpu_present and indicates CPUs which are available for scheduling. These masks depend on the
CONFIG_HOTPLUG_CPU configuration option and if this option is disabled possible == present and active == online .
Implementation of the all of these functions are very similar. Every function checks the second parameter. If it is true , it
calls cpumask_set_cpu or cpumask_clear_cpu otherwise.
For example let's look at set_cpu_possible . As we passed true as the second parameter, the:
cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits));
will be called. First of all let's try to understand the to_cpu_mask macro. This macro casts a bitmap to a struct cpumask * .
CPU masks provide a bitmap suitable for representing the set of CPU's in a system, one bit position per CPU number. CPU
mask presented by the cpu_mask structure:
As we can see from its definition, the DECLARE_BITMAP macro expands to the array of unsigned long . Now let's look at how
the to_cpumask macro is implemented:
#define to_cpumask(bitmap) \
((struct cpumask *)(1 ? (bitmap) \
94
Linux Inside
: (void *)sizeof(__check_is_bitmap(bitmap))))
I don't know about you, but it looked really weird for me at the first time. We can see a ternary operator here which is true
every time, but why the __check_is_bitmap here? It's simple, let's look at it:
Yeah, it just returns 1 every time. Actually we need in it here only for one purpose: at compile time it checks that the given
bitmap is a bitmap, or in other words it checks that the given bitmap has a type of unsigned long * . So we just pass
cpu_possible_bits to the to_cpumask macro for converting the array of unsigned long to the struct cpumask * . Now we
can call cpumask_set_cpu function with the cpu - 0 and struct cpumask *cpu_possible_bits . This function makes only one
call of the set_bit function which sets the given cpu in the cpumask. All of these set_cpu_* functions work on the same
principle.
If you're not sure that this set_cpu_* operations and cpumask are not clear for you, don't worry about it. You can get more
info by reading the special part about it - cpumask or documentation.
As we activated the bootstrap processor, it's time to go to the next function in the start_kernel. Now it is
page_address_init , but this function does nothing in our case, because it executes only when all RAM can't be mapped
directly.
as you can see it just expands to the printk call. At this moment we use pr_notice to print the Linux banner:
pr_notice("%s", linux_banner);
Linux version 4.0.0-rc6+ (alex@localhost) (gcc version 4.9.1 (Ubuntu 4.9.1-16ubuntu6) ) #319 SMP
95
Linux Inside
(you can remember it from the arch/x86/kernel/head_64.S) and ends before __bss_stop . We are using memblock for the
reserving of memory block:
You can read about memblock in the Linux kernel memory management Part 1.. As you can remember memblock_reserve
function takes two parameters:
base physical address of a memory block;
size of a memory block.
We can get the base physical address of the _text symbol with the __pa_symbol macro:
#define __pa_symbol(x) \
__phys_addr_symbol(__phys_reloc_hide((unsigned long)(x)))
First of all it calls __phys_reloc_hide macro on the given parameter. The __phys_reloc_hide macro does nothing for x86_64
and just returns the given parameter. Implementation of the __phys_addr_symbol macro is easy. It just subtracts the symbol
address from the base address of the kernel text mapping base virtual address (you can remember that it is
__START_KERNEL_map ) and adds phys_base which is the base address of _text :
#define __phys_addr_symbol(x) \
((unsigned long)(x) - __START_KERNEL_map + phys_base)
After we got the physical address of the _text symbol, memblock_reserve can reserve a memory block from the _text to
the __bss_stop - _text .
All of these parameters are taken from boot_params . If you have read the chapter about Linux Kernel Booting Process, you
must remember that we filled the boot_params structure during boot time. The kernel setup header contains a couple of
fields which describes ramdisk, for example:
96
Linux Inside
So we can get all the information that interests us from boot_params . For example let's look at get_ramdisk_image :
Here we get the address of the ramdisk from the boot_params and shift left it on 32 . We need to do it because as you can
read in the Documentation/x86/zero-page.txt:
So after shifting it on 32, we're getting a 64-bit address in ramdisk_image and we return it. get_ramdisk_size works on the
same principle as get_ramdisk_image , but it used ext_ramdisk_size instead of ext_ramdisk_image . After we got ramdisk's
size, base address and end address, we check that bootloader provided ramdisk with the:
if (!boot_params.hdr.type_of_loader ||
!ramdisk_image || !ramdisk_size)
return;
and reserve memory block with the calculated addresses for the initial ramdisk in the end:
Conclusion
It is the end of the fourth part about the Linux kernel initialization process. We started to dive in the kernel generic code
from the start_kernel function in this part and stopped on the architecture-specific initializations in the setup_arch . In the
next part we will continue with architecture-dependent initialization steps.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me a PR to linux-internals.
Links
GCC function attributes
this_cpu operations
cpumask
lock validator
cgroups
stack buffer overflow
IRQs
initrd
Previous part
97
Linux Inside
98
Linux Inside
Debug interrupt #DB is the primary means of invoking debuggers. early_trap_init defined in the arch/x86/kernel/traps.c.
This functions sets #DB and #BP handlers and reloads IDT:
We already saw implementation of the set_intr_gate in the previous part about interrupts. Here are two similar functions
set_intr_gate_ist and set_system_intr_gate_ist . Both of these two functions take two parameters:
Both of these functions checks interrupt number and call _set_gate inside:
as set_intr_gate does this. But set_intr_gate calls _set_gate with dpl - 0, and ist - 0, but set_intr_gate_ist and
set_system_intr_gate_ist sets ist as DEBUG_STACK and set_system_intr_gate_ist sets dpl as 0x3 which is the lowest
99
Linux Inside
privilege. When an interrupt occurs and the hardware loads such a descriptor, then hardware automatically sets the new
stack pointer based on the IST value, then invokes the interrupt handler. All of the special kernel stacks will be setted in the
cpu_init function (we will see it later).
As #DB and #BP gates written to the idt_descr , we reload IDT table with load_idt which just cals ldtr instruction. Now
let's look on interrupt handlers and will try to understand how they works. Of course, I can't cover all interrupt handlers in
this book and I do not see the point in this. It is very interesting to delve in the linux kernel source code, so we will see how
debug handler implemented in this part, and understand how other interrupt handlers are implemented will be your task.
DB handler
As you can read above, we passed address of the #DB handler as &debug in the set_intr_gate_ist . lxr.free-electorns.com
is a great resource for searching identificators in the linux kernel source code, but unfortunately you will not find debug
handler with it. All of you can find, it is debug definition in the arch/x86/include/asm/traps.h:
We can see asmlinkage attribute which tells to us that debug is function written with assembly. Yeah, again and again
assembly :). Implementation of the #DB handler as other handlers is in this arch/x86/kernel/entry_64.S and defined with the
idtentry assembly macro:
idtentry is a macro which defines an interrupt/exception entry point. As you can see it takes five arguments:
100
Linux Inside
...
...
You can remember from the previous part about early interrupts/exceptions handling that after interrupt occurs, current
stack will have following format:
+-----------------------+
| |
+40 | SS |
+32 | RSP |
+24 | RFLAGS |
+16 | CS |
+8 | RIP |
0 | Error Code | <---- rsp
| |
+-----------------------+
ASM_CLAC
PARAVIRT_ADJUST_EXCEPTION_FRAME
First ASM_CLAC macro depends on CONFIG_X86_SMAP configuration option and need for security resason, more about it you
can read here. The second PARAVIRT_ADJUST_EXCEPTION_FRAME macro is for handling handle Xen-type-exceptions (this
chapter about kernel initializations and we will not consider virtualization stuff here).
The next piece of code checks is interrupt has error code or not and pushes $-1 which is 0xffffffffffffffff on x86_64
on the stack if not:
.ifeq \has_error_code
pushq_cfi $-1
.endif
We need to do it as dummy error code for stack consistency for all interrupts. In the next step we subscract from the stack
pointer $ORIG_RAX-R15 :
where ORIRG_RAX , R15 and other macros defined in the arch/x86/include/asm/calling.h and ORIG_RAX-R15 is 120 bytes.
General purpose registers will occupy these 120 bytes because we need to store all registers on the stack during interrupt
handling. After we set stack for general purpose registers, the next step is checking that interrupt came from userspace
with:
Here we checks first and second bits in the CS . You can remember that CS register contains segment selector where first
two bits are RPL . All privilege levels are integers in the range 03, where the lowest number corresponds to the highest
privilege. So if interrupt came from the kernel mode we call save_paranoid or jump on label 1 if not. In the save_paranoid
we store all general purpose registers on the stack and switch user gs on kernel gs if need:
movl $1,%ebx
101
Linux Inside
movl $MSR_GS_BASE,%ecx
rdmsr
testl %edx,%edx
js 1f
SWAPGS
xorl %ebx,%ebx
1: ret
In the next steps we put pt_regs pointer to the rdi , save error code in the rsi if it is and call interrupt handler which is do_debug in our case from the arch/x86/kernel/traps.c. do_debug like other handlers takes two parameters:
pt_regs - is a structure which presents set of CPU registers which are saved in the process' memory region;
error code - error code of interrupt.
After interrupt handler finished its work, calls paranoid_exit which restores stack, switch on userspace if interrupt came
from there and calls iret . That's all. Of course it is not all :), but we will see more deeply in the separate chapter about
interrupts.
This is general view of the idtentry macro for #DB interrupt. All interrupts are similar on this implementation and defined
with idtentry too. After early_trap_init finished its work, the next function is early_cpu_init . This function defined in the
arch/x86/kernel/cpu/common.c and collects information about a CPU and its vendor.
pmd_t *pmd;
BUILD_BUG_ON((fix_to_virt(0) + PAGE_SIZE) & ((1 << PMD_SHIFT) - 1));
fixmap - is fixed virtual address mappings which extends from FIXADDR_START to FIXADDR_TOP . Fixed virtual addresses are
needed for subsystems that need to know the virtual address at compile time. After the check early_ioremap_init makes a
call of the early_ioremap_setup function from the mm/early_ioremap.c. early_ioremap_setup fills slot_virt arry of the
unsigned long with virtual addresses with 512 temporary boot-time fix-mappings:
After this we get page middle directory entry for the FIX_BTMAP_BEGIN and put to the pmd variable, fills with zeros bm_pte
102
Linux Inside
which is boot time page tables and call pmd_populate_kernel function for setting given page table entry in the given page
middle directory:
pmd = early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN));
memset(bm_pte, 0, sizeof(bm_pte));
pmd_populate_kernel(&init_mm, pmd, bm_pte);
That's all for this. If you feeling missunderstanding, don't worry. There is special part about ioremap and fixmaps in the
Linux Kernel Memory Management. Part 2 chapter.
ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
This code obtains major and minor numbers for the root device where initrd will be mounted later in the do_mount_root
function. Major number of the device identifies a driver associated with the device. Minor number referred on the device
controlled by driver. Note that old_decode_dev takes one parameter from the boot_params_structure . As we can read from
the x86 linux kernel boot protocol:
Now let's try understand what is it old_decode_dev . Actually it just calls MKDEV inside which generates dev_t from the give
major and minor numbers. It's implementation pretty easy:
where dev_t is a kernel data type to present major/minor number pair. But what's the strange old_ prefix? For historical
reasons, there are two ways of managing the major and minor numbers of a device. In the first way major and minor
numbers occupied 2 bytes. You can see it in the previous code: 8 bit for major number and 8 bit for minor number. But there
is problem with this way: 256 major numbers and 256 minor numbers are possible. So 16-bit integer was replaced with 32bit integer where 12 bits reserved for major number and 20 bits for minor. You can see this in the new_decode_dev
implementation:
After calculation we will get 0xfff or 12 bits for major if it is 0xffffffff and 0xfffff or 20 bits for minor . So in the end of
Continue architecture-specific boot-time initializations
103
Linux Inside
execution of the old_decode_dev we will get major and minor numbers for the root device in ROOT_DEV .
screen_info = boot_params.screen_info;
edid_info = boot_params.edid_info;
saved_video_mode = boot_params.hdr.vid_mode;
bootloader_type = boot_params.hdr.type_of_loader;
if ((bootloader_type >> 4) == 0xe) {
bootloader_type &= 0xf;
bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
}
bootloader_version = bootloader_type & 0xf;
bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
All of these parameters we got during boot time and stored in the boot_params structure. After this we need to setup the
end of the I/O memory. As you know the one of the main purposes of the kernel is resource management. And one of the
resource is a memory. As we already know there are two ways to communicate with devices are I/O ports and device
memory. All information about registered resources available through:
/proc/ioports - provides a list of currently registered port regions used for input or output communication with a device;
/proc/iomem - provides current map of the system's memory for each physical device.
At the moment we are interested in /proc/iomem :
cat /proc/iomem
00000000-00000fff : reserved
00001000-0009d7ff : System RAM
0009d800-0009ffff : reserved
000a0000-000bffff : PCI Bus 0000:00
000c0000-000cffff : Video ROM
000d0000-000d3fff : PCI Bus 0000:00
000d4000-000d7fff : PCI Bus 0000:00
000d8000-000dbfff : PCI Bus 0000:00
000dc000-000dffff : PCI Bus 0000:00
000e0000-000fffff : reserved
000e0000-000e3fff : PCI Bus 0000:00
000e4000-000e7fff : PCI Bus 0000:00
000f0000-000fffff : System ROM
As you can see range of addresses are shown in hexadecimal notation with its owner. Linux kernel provides API for
managing any resources in a general way. Global resources (for example PICs or I/O ports) can be divided into subsets relating to any hardware bus slot. The main structure resource :
struct resource {
resource_size_t start;
resource_size_t end;
const char *name;
unsigned long flags;
struct resource *parent, *sibling, *child;
};
presents abstraction for a tree-like subset of system resources. This structure provides range of addresses from start to
104
Linux Inside
end ( resource_size_t is phys_addr_t or u64 for x86_64 ) which a resource covers, name of a resource (you see these
names in the /proc/iomem output) and flags of a resource (All resources flags defined in the include/linux/ioport.h). The
last are three pointers to the resource structure. These pointers enable a tree-like structure:
+-------------+ +-------------+
| | | |
| parent |------| sibling |
| | | |
+-------------+ +-------------+
|
|
+-------------+
| |
| child |
| |
+-------------+
Every subset of resources has root range resources. For iomem it is iomem_resource which defined as:
TODO EXPORT_SYMBOL
iomem_resource defines root addresses range for io memory with PCI mem name and IORESOURCE_MEM ( 0x00000200 ) as flags.
As i wrote about our current point is setup the end address of the iomem . We will do it with:
Here we shift 1 on boot_cpu_data.x86_phys_bits . boot_cpu_data is cpuinfo_x86 structure which we filled during execution
of the early_cpu_init . As you can understand from the name of the x86_phys_bits field, it presents maximum bits amount
of the maximum physical address in the system. Note also that iomem_resource passed to the EXPORT_SYMBOL macro. This
macro exports the given symbol ( iomem_resource in our case) for dynamic linking or in another words it makes a symbol
accessible to dynamically loaded modules.
As we set the end address of the root iomem resource address range, as I wrote about the next step will be setup of the
memory map. It will be produced with the call of the setup_memory_map function:
First of all we call look here the call of the x86_init.resources.memory_setup . x86_init is a x86_init_ops structure which
presents platform specific setup functions as resources initializtion, pci initialization and etc... Initiaization of the x86_init is
in the arch/x86/kernel/x86_init.c. I will not give here the full description because it is very long, but only one part which
interests us for now:
Continue architecture-specific boot-time initializations
105
Linux Inside
As we can see here memry_setup field is default_machine_specific_memory_setup where we get the number of the e820
entries which we collected in the boot time, sanitize the BIOS e820 map and fill e820map structure with the memory regions.
As all regions collect, print of all regions with printk. You can find this print if you execute dmesg command, you must see
something like this:
It used for storing setup information for different types as device tree blob, EFI setup data and etc... In the second step we
copy BIOS EDD informantion from the boot_params structure that we collected in the arch/x86/boot/edd.c to the edd
structure:
106
Linux Inside
#define INIT_TASK(tsk) \
{
...
...
...
.mm = NULL, \
.active_mm = &init_mm, \
...
}
mm points to the process address space and active_mm points to the active address space if process has no own as kernel
threads (more about it you can read in the documentation). Now we fill memory descriptor of the initial process:
with the kernel's text, data and brk. init_mm is memory descriptor of the initial process and defined as:
where mm_rb is a red-black tree of the virtual memory areas, pgd is a pointer to the page global directory, mm_users is
address space users, mm_count is primary usage counter and mmap_sem is memory area semaphore. After that we setup
memory descriptor of the initiali process, next step is initialization of the intel Memory Protection Extensions with
mpx_mm_init . The next step after it is initialization of the code/data/bss resources with:
code_resource.start = __pa_symbol(_text);
code_resource.end = __pa_symbol(_etext)-1;
data_resource.start = __pa_symbol(_etext);
data_resource.end = __pa_symbol(_edata)-1;
bss_resource.start = __pa_symbol(__bss_start);
bss_resource.end = __pa_symbol(__bss_stop)-1;
We already know a little about resource structure (read above). Here we fills code/data/bss resources with the physical
addresses of they. You can see it in the /proc/iomem output:
Continue architecture-specific boot-time initializations
107
Linux Inside
All of these structures defined in the arch/x86/kernel/setup.c and look like typical resource initialization:
The last step which we will cover in this part will be NX configuration. NX-bit or no execute bit is 63-bit in the page
directory entry which controls the ability to execute code from all physical pages mapped by the table entry. This bit can
only be used/set when the no-execute page-protection mechanism is enabled by the setting EFER.NXE to 1. In the
x86_configure_nx function we check that CPU has support of NX-bit and it does not disabled. After the check we fill
__supported_pte_mask depend on it:
void x86_configure_nx(void)
{
if (cpu_has_nx && !disable_nx)
__supported_pte_mask |= _PAGE_NX;
else
__supported_pte_mask &= ~_PAGE_NX;
}
Conclusion
It is the end of the fifth part about linux kernel initialization process. In this part we continued to dive in the setup_arch
function which makes initialization of architecutre-specific stuff. It was long part, but we not finished with it. As i already
wrote, the setup_arch is big function, and I am really not sure that we will cover full of it even in the next part. There were
some new interesting concepts in this part like Fix-mapped addresses, ioremap and etc... Don't worry if they are unclear for
you. There is special part about these concepts - Linux kernel memory management Part 2.. In the next part we will
continue with the initialization of the architecture-specific stuff and will see parsing of the early kernel parameteres, early
dump of the pci devices, direct Media Interface scanning and many many more.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
mm vs active_mm
e820
Supervisor mode access prevention
Kernel stacks
TSS
IDT
Memory mapped I/O
CFI directives
Continue architecture-specific boot-time initializations
108
Linux Inside
109
Linux Inside
architecture-specific initialization process. The next function after x86_configure_nx is parse_early_param . This function
defined in the init/main.c and as you can understand from its name, this function parses kernel command line and setups
different some services depends on give parameters (all kernel command line parameters you can find in the
Documentation/kernel-parameters.txt). You can remember how we setup earlyprintk in the earliest part. On the early
stage we looked for kernel parameters and their value with the cmdline_find_option function and __cmdline_find_option ,
__cmdline_find_option_bool helpers from the arch/x86/boot/cmdline.c. There we're in the generic kernel part which does
not depend on architecture and here we use another approach. If you are reading linux kernel source code, you already
can note calls like this:
early_param("gbpages", parse_direct_gbpages_on);
in the include/linux/init.h. As you can see early_param macro just makes call of the __setup_param macro:
This macro defines __setup_str_*_id variable (where * depends on given function name) and assigns it to the given
command line parameter name. In the next line we can see definition of the __setup_* variable which type is
obs_kernel_param and its initialization. obs_kernel_param structure defined as:
struct obs_kernel_param {
const char *str;
int (*setup_func)(char *);
int early;
};
110
Linux Inside
#define INIT_SETUP(initsetup_align) \
. = ALIGN(initsetup_align); \
VMLINUX_SYMBOL(__setup_start) = .; \
*(.init.setup) \
VMLINUX_SYMBOL(__setup_end) = .;
Now we know how parameters are defined, let's back to the parse_early_param implementation:
The parse_early_param function defines two static variables. First done check that parse_early_param already called and
the second is temporary storage for kernel command line. After this we copy boot_command_line to the temporary commad
line which we just defined and call the parse_early_options function from the the same source code main.c file.
parse_early_options calls the parse_args function from the kernel/params.c where parse_args parses given command line
and calls do_early_param function. This function goes from the __setup_start to __setup_end , and calls the function from
the obs_kernel_param if a parameter is early. After this all services which are depend on early command line parameters
were setup and the next call after the parse_early_param is x86_report_nx . As I wrote in the beginning of this part, we
already set NX-bit with the x86_configure_nx . The next x86_report_nx function the arch/x86/mm/setup_nx.c just prints
information about the NX . Note that we call x86_report_nx not right after the x86_configure_nx , but after the call of the
parse_early_param . The answer is simple: we call it after the parse_early_param because the kernel support noexec
parameter:
noexec [X86]
On X86-32 available only on PAE configured kernels.
noexec=on: enable non-executable mappings (default)
noexec=off: disable non-executable mappings
memblock_x86_reserve_range_setup_data();
111
Linux Inside
function. This function defined in the same arch/x86/kernel/setup.c source code file and remaps memory for the setup_data
and reserved memory block for the setup_data (more about setup_data you can read in the previous part and about
ioremap and memblock you can read in the Linux kernel memory management).
if (acpi_mps_check()) {
#ifdef CONFIG_X86_LOCAL_APIC
disable_apic = 1;
#endif
setup_clear_cpu_cap(X86_FEATURE_APIC);
}
The first acpi_mps_check function from the arch/x86/kernel/acpi/boot.c depends on CONFIG_X86_LOCAL_APIC and
CNOFIG_x86_MPPARSE configuration options:
It checks the built-in MPS or MultiProcessor Specification table. If CONFIG_X86_LOCAL_APIC is set and CONFIG_x86_MPPAARSE is
not set, acpi_mps_check prints warning message if the one of the command line options: acpi=off , acpi=noirq or
pci=noacpi passed to the kernel. If acpi_mps_check returns 1 which means that
we disable local APIC and clears X86_FEATURE_APIC bit in the of the current CPU with the setup_clear_cpu_cap macro. (more
about CPU mask you can read in the CPU masks).
#ifdef CONFIG_PCI
if (pci_early_dump_regs)
early_dump_pci_devices();
#endif
pci_early_dump_regs variable defined in the arch/x86/pci/common.c and its value depends on the kernel command line
early_param("pci", pci_setup);
pci_setup function gets the string after the pci= and analyzes it. This function calls pcibios_setup which defined as
__weak in the drivers/pci/pci.c and every architecture defines the same function which overrides __weak analog. For
112
Linux Inside
So, if CONFIG_PCI option is set and we passed pci=earlydump option to the kernel command line, next function which will be
called - early_dump_pci_devices from the arch/x86/pci/early.c. This function checks noearly pci parameter with:
if (!early_pci_allowed())
return;
and returns if it was passed. Each PCI domain can host up to 256 buses and each bus hosts up to 32 devices. So, we
goes in a loop:
Let's look on it. As you can see the first function is e820_reserve_setup_data . This function does almost the same as
memblock_x86_reserve_range_setup_data which we saw above, but it also calls e820_update_range which adds new regions to
the e820map with the given type which is E820_RESERVED_KERN in our case. The next function is finish_e820_parsing which
sanitazes e820map with the sanitize_e820_map function. Besides this two functions we can see a couple of functions related
to the e820. You can see it in the listing which is above. e820_add_kernel_range function takes the physical address of the
Architecture-specific initializations, again...
113
Linux Inside
checks that .text .data and .bss marked as E820RAM in the e820map and prints the warning message if not. The next
function trm_bios_range update first 4096 bytes in e820Map as E820_RESERVED and sanitizes it again with the call of the
sanitize_e820_map . After this we get the last page frame number with the call of the e820_end_of_ram_pfn function. Every
memory page has an unique number - Page frame number and e820_end_of_ram_pfn function returns the maximum with the
call of the e820_end_pfn :
where e820_end_pfn takes maximum page frame number on the certain architecture ( MAX_ARCH_PFN is 0x400000000 for
x86_64 ). In the e820_end_pfn we go through the all e820 slots and check that e820 entry has E820_RAM or E820_PRAM type
because we calcluate page frame numbers only for these types, gets the base address and end address of the page frame
number for the current e820 entry and makes some checks for these addresses:
After this we check that last_pfn which we got in the loop is not greater that maximum page frame number for the certain
architecture ( x86_64 in our case), print inofmration about last page frame number and return it. We can see the last_pfn in
the dmesg output:
...
[ 0.000000] e820: last_pfn = 0x41f000 max_arch_pfn = 0x400000000
...
114
Linux Inside
After this, as we have calculated the biggest page frame number, we calculate max_low_pfn which is the biggest page
frame number in the low memory or bellow first 4 gigabytes. If installed more than 4 gigabytes of RAM, max_low_pfn will be
result of the e820_end_of_low_ram_pfn function which does the same e820_end_of_ram_pfn but with 4 gigabytes limit, in other
way max_low_pfn will be the same as max_pfn :
Next we calculate high_memory (defines the upper bound on direct map memory) with __va macro which returns a virtual
address by the given physical.
DMI scanning
The next step after manipulations with different memory regions and e820 slots is collecting information about computer.
We will get all information with the Desktop Management Interface and following functions:
dmi_scan_machine();
dmi_memdev_walk();
First is dmi_scan_machine defined in the drivers/firmware/dmi_scan.c. This function goes through the System Management
BIOS structures and extracts informantion. There are two ways specified to gain access to the SMBIOS table: get the pointer
to the SMBIOS table from the EFI's configuration table and scanning the physycal memory between 0xF0000 and 0x10000
addresses. Let's look on the second approach. dmi_scan_machine function remaps memory between 0xf0000 and 0x10000
with the dmi_early_remap which just expands to the early_ioremap :
and iterates over all DMI header address and find search _SM_ string:
memset(buf, 0, 16);
for (q = p; q < p + 0x10000; q += 16) {
memcpy_fromio(buf + 16, q, 16);
if (!dmi_smbios3_present(buf) || !dmi_present(buf)) {
dmi_available = 1;
dmi_early_unmap(p, 0x10000);
goto out;
}
memcpy(buf, buf + 16, 16);
}
_SM_ string must be between 000F0000h and 0x000FFFFF . Here we copy 16 bytes to the buf with memcpy_fromio which is
the same memcpy and execute dmi_smbios3_present and dmi_present on the buffer. These functions check that first 4 bytes
is _SM_ string, get SMBIOS version and gets _DMI_ attributes as DMI structure table length, table address and etc... After
Architecture-specific initializations, again...
115
Linux Inside
one of these function will finish to execute, you will see the result of it in the dmesg output:
dmi_early_unmap(p, 0x10000);
The second function is - dmi_memdev_walk . As you can understand it goes over memory devices. Let's look on it:
It checks that DMI available (we got it in the previous function - dmi_scan_machine ) and collects information about memory
devices with dmi_walk_early and dmi_alloc which defined as:
#ifdef CONFIG_DMI
RESERVE_BRK(dmi_alloc, 65536);
#endif
RESERVE_BRK defined in the arch/x86/include/asm/setup.h and reserves space with given size in the brk section.
init_hypervisor_platform();
x86_init.resources.probe_roms();
insert_resource(&iomem_resource, &code_resource);
insert_resource(&iomem_resource, &data_resource);
insert_resource(&iomem_resource, &bss_resource);
early_gart_iommu_check();
SMP config
The next step is parsing of the SMP configuration. We do it with the call of the find_smp_config function which just calls
function:
116
Linux Inside
if (smp_scan_config(0x0, 0x400) ||
smp_scan_config(639 * 0x400, 0x400) ||
smp_scan_config(0xF0000, 0x10000))
return;
First is virtual address of the memory region where we will scan SMP config, second is the pointer to the mpf_intel
structure. Let's try to understand what is it mpf_intel . All information stores in the multiprocessor configuration data
structure. mpf_intel presents this structure and looks:
struct mpf_intel {
char signature[4];
unsigned int physptr;
unsigned char length;
unsigned char specification;
unsigned char checksum;
unsigned char feature1;
unsigned char feature2;
unsigned char feature3;
unsigned char feature4;
unsigned char feature5;
};
As we can read in the documentation - one of the main functions of the system BIOS is to construct the MP floating pointer
structure and the MP configuration table. And operating system must have access to this information about the
multiprocessor configuration and mpf_intel stores the physical address (look at second parameter) of the multiprocessor
configuration table. So, smp_scan_config going in a loop through the given memory range and tries to find MP floating
pointer structure there. It checks that current byte points to the SMP signature, checks checksum, checks that mpf>specification is 1 (it must be 1 or 4 by specification) in the loop:
reserves given memory block if search is successful with memblock_reserve and reserves physical address of the
multiprocessor configuration table. All documentation about this you can find in the - MultiProcessor Specification. More
details you can read in the special part about SMP .
117
Linux Inside
First of all it get the size of the page table buffer, it will be INIT_PGT_BUF_SIZE which is (6 * PAGE_SIZE) in the current linux
kernel 4.0. As we got the size of the page table buffer, we call extend_brk function with two parameters: size and align. As
you can understand from its name, this function extends the brk area. As we can see in the linux kernel linker script brk
in memory right after the BSS:
. = ALIGN(PAGE_SIZE);
.brk : AT(ADDR(.brk) - LOAD_OFFSET) {
__brk_base = .;
. += 64 * 1024; /* 64k alignment slop space */
*(.brk_reservation) /* areas brk users have reserved */
__brk_limit = .;
}
After that we got physical address of the new brk with the __pa macro, we calculate the base address and the end of the
page table buffer. In the next step as we got page table buffer, we reserve memory block for the brk are with the
reserve_brk function:
Note that in the end of the reserve_brk , we set brk_start to zero, because after this we will not allocate it anymore. The
next step after reserving memory block for the brk , we need to unmap out-of-range memory areas in the kernel mapping
with the cleanup_highmap function. Remeber that kernel mapping is __START_KERNEL_map and _end - _text or
level2_kernel_pgt maps the kernel _text , data and bss . In the start of the clean_high_map we define these parameters:
Now, as we defined start and end of the kernel mapping, we go in the loop through the all kernel page middle directory
entries and clean entries which are not between _text and end :
Architecture-specific initializations, again...
118
Linux Inside
After this we set the limit for the memblock allocation with the memblock_set_current_limit function (read more about
memblock you can in the Linux kernel memory management Part 2), it will be ISA_END_ADDRESS or 0x100000 and fill the
memblock information according to e820 with the call of the memblock_x86_fill function. You can see the result of this
MEMBLOCK configuration:
memory size = 0x1fff7ec00 reserved size = 0x1e30000
memory.cnt = 0x3
memory[0x0] [0x00000000001000-0x0000000009efff], 0x9e000 bytes flags: 0x0
memory[0x1] [0x00000000100000-0x000000bffdffff], 0xbfee0000 bytes flags: 0x0
memory[0x2] [0x00000100000000-0x0000023fffffff], 0x140000000 bytes flags: 0x0
reserved.cnt = 0x3
reserved[0x0] [0x0000000009f000-0x000000000fffff], 0x61000 bytes flags: 0x0
reserved[0x1] [0x00000001000000-0x00000001a57fff], 0xa58000 bytes flags: 0x0
reserved[0x2] [0x0000007ec89000-0x0000007fffffff], 0x1377000 bytes flags: 0x0
The rest functions after the memblock_x86_fill are: early_reserve_e820_mpc_new alocates additional slots in the e820map for
MultiProcessor Specification table, reserve_real_mode - reserves low memory from 0x0 to 1 megabyte for the trampoline to
the real mode (for rebootin and etc...), trim_platform_memory_ranges - trims certain memory regions started from
0x20050000 , 0x20110000 and etc... these regions must be excluded because Sandy Bridge has problems with these
regions, trim_low_memory_range reserves the first 4 killobytes page in memblock , init_mem_mapping function reconstructs
direct memory mapping and setups the direct mapping of the physical memory at PAGE_OFFSET , early_trap_pf_init setups
#PF handler (we will look on it in the chapter about interrupts) and setup_real_mode function setups trampoline to the real
mode code.
That's all. You can note that this part will not cover all functions which are in the setup_arch (like early_gart_iommu_check ,
mtrr initalization and etc...). As I already wrote many times, setup_arch is big, and linux kernel is big. That's why I can't
cover every line in the linux kernel. I don't think that we missed something important,... but you can say something like:
each line of code is important. Yes, it's true, but I missed they anyway, because I think that it is not real to cover full linux
kernel. Anyway we will often return to the idea that we have already seen, and if something will be unfamiliar, we will cover
this theme.
Conclusion
It is the end of the sixth part about linux kernel initialization process. In this part we continued to dive in the setup_arch
function again It was long part, but we not finished with it. Yes, setup_arch is big, hope that next part will be last about this
function.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
MultiProcessor Specification
NX bit
Architecture-specific initializations, again...
119
Linux Inside
Documentation/kernel-parameters.txt
APIC
CPU masks
Linux kernel memory management
PCI
e820
System Management BIOS
System Management BIOS
EFI
SMP
MultiProcessor Specification
BSS
SMBIOS specification
Previous part
120
Linux Inside
Now let's look on the implementation of th setup_log_buf function. It starts with check that current buffer is empty (It must
be empty, because we just setup it) and another check that it is early setup. If setup of the kernel log buffer is not early, we
call the log_buf_add_cpu function which increase size of the buffer for every CPU:
if (log_buf != __log_buf)
return;
if (!early && !new_log_buf_len)
log_buf_add_cpu();
We will not research log_buf_add_cpu function, because as you can see in the setup_arch , we call setup_log_buf as:
setup_log_buf(1);
where 1 means that is is early setup. In the next step we check new_log_buf_len variable which is updated length of the
kernel log buffer and allocate new space for the buffer with the memblock_virt_alloc function for it, or just return.
As kernel log buffer is ready, the next function is reserve_initrd . You can remember that we already called the
early_reserve_initrd function in the fourth part of the Kernel initialization. Now, as we reconstructed direct memory
mapping in the init_mem_mapping function, we need to move initrd to the down into directly mapped memory. The
reserve_initrd function starts from the definition of the base address and end address of the initrd and check that
initrd was provided by a bootloader. All the same as we saw it in the early_reserve_initrd . But instead of the reserving
place in the memblock area with the call of the memblock_reserve function, we get the mapped size of the direct memory
area and check that the size of the initrd is not greater that this area with:
mapped_size = memblock_mem_size(max_pfn_mapped);
if (ramdisk_size >= (mapped_size>>1))
panic("initrd too large to handle, "
"disabling initrd (%lld needed, %lld available)\n",
121
Linux Inside
ramdisk_size, mapped_size>>1);
You can see here that we call memblock_mem_size function and pass the max_pfn_mapped to it, where max_pfn_mapped
contains the highest direct mapped page frame number. If you do not remember what is it page frame number , explanation
is simple: First 12 bits of the virtual address represent offset in the physical page or page frame. If we will shift right virtual
address on 12 , we'll discard offset part and will get Page Frame Number . In the memblock_mem_size we go through the all
memblock mem (not reserved) regions and calculates size of the mapped pages amount and return it to the mapped_size
variable (see code above). As we got amount of the direct mapped memory, we check that size of the initrd is not greater
than mapped pages. If it is greater we just call panic which halts the system and prints popular Kernel panic message. In
the next step we print information about the initrd size. We can see the result of this in the dmesg output:
and relocate initrd to the direct mapping area with the relocate_initrd function. In the start of the relocate_initrd
function we try to find free area with the memblock_find_in_range function:
The memblock_find_in_range function tries to find free area in a given range, in our case from 0 to the maximum mapped
physical address and size must equal to the aligned size of the initrd . If we didn't find area with the given size, we call
panic again. If all is good, we start to relocated RAM disk to the down of the directly mapped meory in the next step.
In the end of the reserve_initrd function, we free memblock memory which occupied by the ramdisk with the call of the:
After we relocated initrd ramdisk image, the next function is vsmp_init from the arch/x86/kernel/vsmp_64.c. This
function initializes support of the ScaleMP vSMP . As I already wrote in the previous parts, this chapter will not cover nonrelated x86_64 initialization parts (for example as the current or ACPI and etc...). So we will miss implementation of this for
now and will back to it in the part which will cover techniques of parallel computing.
The next function is io_delay_init from the arch/x86/kernel/io_delay.c. This function allows to override default default I/O
delay 0x80 port. We already saw I/O delay in the Last preparation before transition into protected mode, now let's look on
the io_delay_init implementation:
This function check io_delay_override variable and overrides I/O delay port if io_delay_override is set. We can set
io_delay_override variably by passing io_delay option to the kernel command line. As we can read from the
122
Linux Inside
We can see io_delay command line parameter setup with the early_param macro in the arch/x86/kernel/io_delay.c
early_param("io_delay", io_delay_param);
More about early_param you can read in the previous part. So the io_delay_param function which setups
io_delay_override variable will be called in the do_early_param function. io_delay_param function gets the argument of the
io_delay kernel command line parameter and sets io_delay_type depends on it:
The next functions are acpi_boot_table_init , early_acpi_boot_init and initmem_init after the io_delay_init , but as I
wrote above we will not cover ACPI related stuff in this Linux Kernel initialization process chapter.
phys_addr_t selected_size = 0;
phys_addr_t selected_base = 0;
phys_addr_t selected_limit = limit;
bool fixed = false;
where first represents size in bytes of the reserved area, second is base address of the reserved area, third is end address
of the reserved area and the last fixed parameter shows where to place reserved area. If fixed is 1 we just reserve area
with the memblock_reserve , if it is 0 we allocate space with the kmemleak_alloc . In the next step we check size_cmdline
variable and if it is not equal to -1 we fill all variables which you can see above with the values from the cma kernel
command line parameter:
123
Linux Inside
if (size_cmdline != -1) {
...
...
...
}
You can find in this source code file definition of the early parameter:
early_param("cma", early_cma);
cma=nn[MG]@[start[MG][-end[MG]]]
[ARM,X86,KNL]
Sets the size of kernel global memory area for
contiguous memory allocations and optionally the
placement constraint by the physical address range of
memory allocations. A value of 0 disables CMA
altogether. For more information, see
include/linux/dma-contiguous.h
If we will not pass cma option to the kernel command line, size_cmdline will be equal to -1 . In this way we need to
calculate size of the reserved area which depends on the following kernel configuration options:
CONFIG_CMA_SIZE_SEL_MBYTES - size in megabytes, default global CMA area, which is equal to CMA_SIZE_MBYTES * SZ_1M or
CONFIG_CMA_SIZE_MBYTES * 1M ;
CONFIG_CMA_SIZE_SEL_PERCENTAGE - percentage of total memory;
CONFIG_CMA_SIZE_SEL_MIN - use lower value;
CONFIG_CMA_SIZE_SEL_MAX - use higher value.
As we calculated the size of the reserved area, we reserve area with the call of the dma_contiguous_reserve_area function
which first of all calls:
function. The cma_declare_contiguous reserves contiguous area from the given base address and with given size. After we
reserved area for the DMA , next function is the memblock_find_dma_reserve . As you can understand from its name, this
function counts the reserved pages in the DMA area. This part will not cover all details of the CMA and DMA , because they
are big. We will see much more details in the special part in the Linux Kernel Memory management which covers
contiguous memory allocators and areas.
which expands as you can see to the call of the paging_init function from the arch/x86/mm/init_64.c. The paging_init
function initializes sparse memory and zone sizes. First of all what's zones and what is it Sparsemem . The Sparsemem is a
special foundation in the linux kernen memory manager which used to split memory area to the different memory banks in
End of the architecture-specific initializations, almost...
124
Linux Inside
the NUMA systems. Let's look on the implementation of the paginig_init function:
As you can see there is call of the sparse_memory_present_with_active_regions function which records a memory area for
every NUMA node to the array of the mem_section structure which contains a pointer to the structure of the array of struct
page . The next sparse_init function allocates non-linear mem_section and mem_map . In the next step we clear state of the
movable memory nodes and initialize sizes of zones. Every NUMA node is devided into a number of pieces which are called
- zones . So, zone_sizes_init function from the arch/x86/mm/init.c initializes size of zones.
Again, this part and next parts do not cover this theme in full details. There will be special part about NUMA .
vsyscall mapping
The next step after SparseMem initialization is setting of the trampoline_cr4_features which must contain content of the cr4
Control register. First of all we need to check that current CPU has support of the cr4 register and if it has, we save its
content to the trampoline_cr4_features which is storage for cr4 in the real mode:
if (boot_cpu_data.cpuid_level >= 0) {
mmu_cr4_features = __read_cr4();
if (trampoline_cr4_features)
*trampoline_cr4_features = mmu_cr4_features;
}
The next function which you can see is map_vsyscal from the arch/x86/kernel/vsyscall_64.c. This function maps memory
space for vsyscalls and depends on CONFIG_X86_VSYSCALL_EMULATION kernel configuration option. Actually vsyscall is a
special segment which provides fast access to the certain system calls like getcpu and etc... Let's look on implementation
of this function:
In the beginning of the map_vsyscal we can see definition of two variables. The first is extern valirable __vsyscall_page . As
variable extern, it defined somewhere in other source code file. Actually we can see definition of the __vsyscall_page in the
arch/x86/kernel/vsyscall_emu_64.S. The __vsyscall_page symbol points to the aligned calls of the vsyscalls as
gettimeofday and etc...:
125
Linux Inside
.globl __vsyscall_page
.balign PAGE_SIZE, 0xcc
.type __vsyscall_page, @object
__vsyscall_page:
mov $__NR_gettimeofday, %rax
syscall
ret
.balign 1024, 0xcc
mov $__NR_time, %rax
syscall
ret
...
...
...
The second variable is physaddr_vsyscall which just stores physical address of the __vsyscall_page symbol. In the next
step we check the vsyscall_mode variable, and if it is not equal to NONE which is EMULATE by default:
And after this check we can see the call of the __set_fixmap function which calls native_set_fixmap with the same
parameters:
Here we can see that native_set_fixmap makes value of Page Table Entry from the given physical address (physical
address of the __vsyscall_page symbol in our case) and calls internal function - __native_set_fixmap . Internal function gets
the virtual address of the given fixed_addresses index ( VSYSCALL_PAGE in our case) and checks that given index is not
greated than end of the fix-mapped addresses. After this we set page table entry with the call of the set_pte_vaddr function
and increase count of the fix-mapped addresses. And in the end of the map_vsyscall we check that virtual address of the
VSYSCALL_PAGE (which is first index in the fixed_addresses ) is not greater than VSYSCALL_ADDR which is -10UL << 20 or
ffffffffff600000 with the BUILD_BUG_ON macro:
BUILD_BUG_ON((unsigned long)__fix_to_virt(VSYSCALL_PAGE) !=
(unsigned long)VSYSCALL_ADDR);
Now vsyscall area is in the fix-mapped area. That's all about map_vsyscall , if you do not know anything about fix-mapped
addresses, you can read Fix-Mapped Addresses and ioremap. More about vsyscalls we will see in the vsyscalls and
vdso part.
126
Linux Inside
You can remember how we made a search of the SMP configuration in the previous part. Now we need to get the SMP
configurtaion if we found it. For this we check smp_found_config variable which we set in the smp_scan_config function (read
about it the previous part) and call the get_smp_config function:
if (smp_found_config)
get_smp_config();
Here we can see that multiprocessor configuration was found in the smp_scan_config function or just return from the
function if not. The next check check that it is early. And as we did this checks, we start to read the SMP configuration. As
we finished to read it, the next step is - prefill_possible_map function which makes preliminary filling of the possible CPUs
cpumask (more about it you can read in the Introduction to the cpumasks).
address of the local APIC. The next is x86_io_apic_ops.init and this function initializes I/O APIC. Please note that all
details related with APIC , we will see in the chapter about interrupts and exceptions handling. In the next step we reserve
standard I/O resources like DMA , TIMER , FPU and etc..., with the call of the x86_init.resources.reserve_resources function.
Following is mcheck_init function initializes Machine check Exception and the last is register_refined_jiffies which
registers jiffy (There will be separate chapter about timers in the kernel).
So that's all. Finally we have finished with the big setup_arch function in this part. Of course as I already wrote many times,
we did not see full details about this function, but do not worry about it. We will be back more than once to this function from
different chapters for understanding how different platform-dependent parts are initialized.
That's all, and now we can back to the start_kernel from the setup_arch .
127
Linux Inside
As you can see in the init/main.c, we passed memory descriptor of the init process to the mm_init_cpumask and here depend
on CONFIG_CPUMASK_OFFSTACK configuration option we set or clear TLB switch cpumask .
In the next step we can see the call of the following function:
setup_command_line(command_line);
This function takes pointer to the kernel command line allocates a couple of buffers to store command line. We need a
couple of buffers, because one buffer used for future reference and accessing to command line and one for parameter
parsing. We will allocate space for the following buffers:
saved_command_line - will contain boot command line;
initcall_command_line - will contain boot command line. will be used in the do_initcall_level ;
static_command_line - will contain command line for parameters parsing.
We will allocate space with the memblock_virt_alloc function. This function calls memblock_virt_alloc_try_nid which
allocates boot memory block with memblock_reserve if slab is not available or uses kzalloc_node (more about it will be in
the linux memory management chapter). The memblock_virt_alloc uses BOOTMEM_LOW_LIMIT (physicall address of the
(PAGE_OFFSET + 0x1000000) value) and BOOTMEM_ALLOC_ACCESSIBLE (equal to the current value of the memblock.current_limit )
as minimum address of the memory egion and maximum address of the memory region.
Let's look on the implementation of the setup_command_line :
Here we can see that we allocate space for the three buffers which will contain kernel command line for the different
purposes (read above). And as we allocated space, we storing boot_comand_line in the saved_command_line and
command_line (kernel command line from the setup_arch to the static_command_line ).
The next function after the setup_command_line is the setup_nr_cpu_ids . This function setting nr_cpu_ids (number of CPUs)
according to the last bit in the cpu_possible_mask (more about it you can read in the chapter describes cpumasks concept).
Let's look on its implementation:
128
Linux Inside
Here nr_cpu_ids represents number of CPUs, NR_CPUS represents the maximum number of CPUs which we can set in
configuration time:
Actually we need to call this function, because NR_CPUS can be greater than actual amount of the CPUs in the your
computer. Here we can see that we call find_last_bit function and pass two parameters to it:
cpu_possible_mask bits;
size to search and returns bit number of the first set bit. We passed cpu_possible_mask bits and maximum number of the
CPUs. First of all the find_last_bit function splits given unsigned long address to the words:
where BITS_PER_LONG is 64 on the x86_64 . As we got amount of words in the given size of the search data, we need to
check is given size does not contain partial words with the following check:
if it contains partial word, we mask the last word and check it. If the last word is not zero, it means that current word
contains at least one set bit. We go to the found label:
129
Linux Inside
found:
return words * BITS_PER_LONG + __fls(tmp);
Here you can see __fls function which returns last set bit in a given word with help of the bsr instruction:
The bsr instruction which scans the given operand for first bit set. If the last word is not partial we going through the all
words in the given address and trying to find first set bit:
while (words) {
tmp = addr[--words];
if (tmp) {
found:
return words * BITS_PER_LONG + __fls(tmp);
}
}
Here we put the last word to the tmp variable and check that tmp contains at least one set bit. If a set bit found, we return
the number of this bit. If no one words do not contains set bit we just return given size:
return size;
After this nr_cpu_ids will contain the correct amount of the avaliable CPUs.
That's all.
Conclusion
It is the end of the seventh part about the linux kernel initialization process. In this part, finally we have finsihed with the
setup_arch function and returned to the start_kernel function. In the next part we will continue to learn generic kernel
code from the start_kernel and will continue our way to the first init process.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
Desktop Management Interface
x86_64
initrd
Kernel panic
Documentation/kernel-parameters.txt
ACPI
End of the architecture-specific initializations, almost...
130
Linux Inside
131
Linux Inside
where the smp_prepare_boot_cpu expands to the call of the native_smp_prepare_boot_cpu function (more about smp_ops will
be in the special parts about SMP ):
The native_smp_prepare_boot_cpu function gets the number of the current CPU (which is Bootstrap processor and its id is
zero) with the smp_processor_id function. I will not explain how the smp_processor_id works, because we alread saw it in
the Kernel entry point part. As we got processor id number we reload Global Descriptor Table for the given CPU with the
switch_to_new_gdt function:
The gdt_descr variable represents pointer to the GDT descriptor here (we already saw desc_ptr in the Early interrupt and
exception handling). We get the address and the size of the GDT descriptor where GDT_SIZE is 256 or:
and the address of the descriptor we will get with the get_cpu_gdt_table :
Scheduler initialization
132
Linux Inside
The get_cpu_gdt_table uses per_cpu macro for getting gdt_page percpu variable for the given CPU number (bootstrap
processor with id - 0 in our case). You can ask the following question: so, if we can access gdt_page percpu variable,
where it was defined? Actually we alread saw it in this book. If you have read the first part of this chapter, you can
remember that we saw definition of the gdt_page in the arch/x86/kernel/head_64.S:
early_gdt_descr:
.word GDT_ENTRIES*8-1
early_gdt_descr_base:
.quad INIT_PER_CPU_VAR(gdt_page)
and if we will look on the linker file we can see that it locates after the __per_cpu_load symbol:
more about percpu variables you can read in the Per-CPU variables part. As we got address and size of the GDT
descriptor we case reload GDT with the load_gdt which just execute lgdt instruct and load percpu_segment with the
following function:
The base address of the percpu area must contain gs register (or fs register for x86 ), so we are using loadsegment
macro and pass gs . In the next step we writes the base address if the IRQ stack and setup stack canary (this is only for
x86_32 ). After we load new GDT , we fill cpu_callout_mask bitmap with the current cpu and set cpu state as online with the
cpumask_set_cpu(me, cpu_callout_mask);
per_cpu(cpu_state, me) = CPU_ONLINE;
So, what is it cpu_callout_mask bitmap... As we initialized bootstrap processor (procesoor which is booted the first on x86 )
the other processors in a multiprocessor system are known as secondary processors . Linux kernel uses two following
bitmasks:
Scheduler initialization
133
Linux Inside
cpu_callout_mask
cpu_callin_mask
After bootstrap processor initialized, it updates the cpu_callout_mask to indicate which secondary processor can be
initialized next. All other or secondary processors can do some initialization stuff before and check the cpu_callout_mask on
the boostrap processor bit. Only after the bootstrap processor filled the cpu_callout_mask this secondary processor, it will
continue the rest of its initialization. After that the certain processor will finish its initialization process, the processor sets bit
in the cpu_callin_mask . Once the bootstrap processor finds the bit in the cpu_callin_mask for the current secondary
processor, this processor repeats the same procedure for initialization of the rest of a secondary processors. In a short
words it works as i described, but more details we will see in the chapter about SMP .
That's all. We did all SMP boot preparation.
Build zonelists
In the next step we can see the call of the build_all_zonelists function. This function sets up the order of zones that
allocations are preferred from. What are zones and what's order we will understand now. For the start let's see how linux
kernel considers physical memory. Physical memory may be arranged into banks which are called - nodes . If you has no
hardware with support for NUMA , you will see only one node:
$ cat /sys/devices/system/node/node0/numastat
numa_hit 72452442
numa_miss 0
numa_foreign 0
interleave_hit 12925
local_node 72452442
other_node 0
Every node presented by the struct pglist data in the linux kernel. Each node devided into a number of special blocks
which are called - zones . Every zone presented by the zone struct in the linux kernel and has one of the type:
ZONE_DMA - 0-16M;
ZONE_DMA32 - used for 32 bit devices that can only do DMA areas below 4G;
ZONE_NORMAL - all RAM from the 4GB on the x86_64 ;
ZONE_HIGHMEM - absent on the x86_64 ;
ZONE_MOVABLE - zone which contains movable pages.
which are presented by the zone_type enum. Information about zones we can get with the:
$ cat /proc/zoneinfo
Node 0, zone DMA
pages free 3975
min 3
low 3
...
...
Node 0, zone DMA32
pages free 694163
min 875
low 1093
...
...
Node 0, zone Normal
pages free 2529995
min 3146
low 3932
...
...
Scheduler initialization
134
Linux Inside
As I wrote above all nodes are described with the pglist_data or pg_data_t structure in memory. This structure defined in
the include/linux/mmzone.h. The build_all_zonelists function from the mm/page_alloc.c constructs an ordered zonelist
(of different zones DMA , DMA32 , NORMAL , HIGH_MEMORY , MOVABLE ) which specifies the zones/nodes to visit when a selected
zone or node cannot satisfy the allocation request. That's all. More about NUMA and multiprocessor systems will be in the
special part.
and initializes handler for the CPU hotplug. Of course the hotcpu_notifier depends on the CONFIG_HOTPLUG_CPU
configuration option and if this option is set, it just calls cpu_notifier macro which expands to the call of the
register_cpu_notifier which adds hotplug cpu handler ( page_alloc_cpu_notify in our case).
After this we can see the kernel command line in the initialization output:
And a couple of functions as parse_early_param and parse_args which are handles linux kernel command line. You can
remember that we already saw the call of the parse_early_param function in the sixth part of the kernel initialization chapter,
so why we call it again? Answer is simple: we call this function in the architecture-specific code ( x86_64 in our case), but
not all architecture calls this function. And we need in the call of the second function parse_args to parse and handle nonearly command line arguments.
In the next step we can see the call of the jump_label_init from the kernel/jump_label.c. and initializes jump label.
After this we can see the call of the setup_log_buf function which setups the printk log buffer. We already saw this function
in the seventh part of the linux kernel initialization process chapter.
value by the kernel. The management of PIDs centered around the two special data structures: struct pid and struct
upid . First structure represents information about a PID in the kernel. The second structure represents the information that
is visible in a specific namespace. All PID instances stored in the special hash table:
This hash table is used to find the pid instance that belongs to a numeric PID value. So, pidhash_init initializes this hash.
In the start of the pidhash_init function we can see the call of the alloc_large_system_hash :
Scheduler initialization
135
Linux Inside
0, 4096);
The number of elements of the pid_hash depends on the RAM configuration, but it can be between 2^4 and 2^12 . The
pidhash_init computes the size and allocates the required storage (which is hlist in our case - the same as doubly
linked list, but contains one pointer instead on the struct hlist_head]. The alloc_large_system_hash function allocates a large
system hash table with memblock_virt_alloc_nopanic if we pass HASH_EARLY flag (as it in our case) or with __vmalloc if we
did no pass this flag.
The result we can see in the dmesg output:
That's all. The rest of the stuff before scheduler initialization is the following functions: vfs_caches_init_early does early
initialization of the virtual file system (more about it will be in the chapter which will describe virtual file system),
sort_main_extable sorts the kernel's built-in exception table entries which are between __start___ex_table and
__stop___ex_table, , and trap_init initializies trap handlers (morea about last two function we will know in the separate
page_ext_init_flatmem();
mem_init();
kmem_cache_init();
percpu_init_late();
pgtable_init();
vmalloc_init();
The first is page_ext_init_flatmem depends on the CONFIG_SPARSEMEM kernel configuration option and initializes extended
data per page handling. The mem_init releases all bootmem , the kmem_cache_init initializes kernel cache, the
percpu_init_late - replaces percpu chunks with those allocated by slub, the pgtable_init - initilizes the vmalloc_init -
initializes vmalloc . Please, NOTE that we will not dive into details about all of these functions and concepts, but we will see
all of they it in the Linux kernem memory manager chapter.
That's all. Now we can look on the scheduler .
Scheduler initialization
And now we came to the main purpose of this part - initialization of the task scheduler. I want to say again as I did it already
many times, you will not see the full explanation of the scheduler here, there will be special chapter about this. Ok, next
point is the sched_init function from the kernel/sched/core.c and as we can understand from the function's name, it
initializes scheduler. Let's start to dive in this function and try to understand how the scheduler initialized. At the start of the
sched_init function we can see the following code:
#ifdef CONFIG_FAIR_GROUP_SCHED
alloc_size += 2 * nr_cpu_ids * sizeof(void **);
#endif
#ifdef CONFIG_RT_GROUP_SCHED
alloc_size += 2 * nr_cpu_ids * sizeof(void **);
#endif
Scheduler initialization
136
Linux Inside
Both of this options provide two different planning models. As we can read from the documentation, the current scheduler CFS or Completely Fair Scheduler used a simple concept. It models process scheduling as if the system had an ideal
multitasking processor where each process would receive 1/n processor time, where n is the number of the runnable
processes. The scheduler uses the special set of rules used. These rules determine when and how to select a new process
to run and they are called scheduling policy . The Completely Fair Scheduler supports following normal or non-real-time
scheduling policies: SCHED_NORMAL , SCHED_BATCH and SCHED_IDLE . The SCHED_NORMAL is used for the most normal
applications, the amount of cpu each process consumes is mostly determined by the nice value, the SCHED_BATCH used for
the 100% non-interactive tasks and the SCHED_IDLE runs tasks only when the processor has not to run anything besides
this task. The real-time policies are also supported for the time-critial applications: SCHED_FIFO and SCHED_RR . If you've
read something about the Linux kernel scheduler, you can know that it is modular. It means that it supports different
algorithms to schedule different types of processes. Usually this modularity is called scheduler classes . These modules
encapsulate scheduling policy details and are handled by the scheduler core without the core code assuming too much
about them.
Now let's back to the our code and look on the two configuration options CONFIG_FAIR_GROUP_SCHED and
CONFIG_RT_GROUP_SCHED . The scheduler operates on an individual task. These options allows to schedule group tasks (more
about it you can read in the CFS group scheduling). We can see that we assign the alloc_size variables which represent
size based on amount of the processors to allocate for the sched_entity and cfs_rq to the 2 * nr_cpu_ids * sizeof(void
**) expression with kzalloc :
The sched_entity is struture which defined in the include/linux/sched.h and used by the scheduler to keep track of process
accounting. The cfs_rq presents run queue. So, you can see that we allocated space with size alloc_size for the run
queue and scheduler entity of the root_task_group . The root_task_group is an instance of the task_group structure from
the kernel/sched/sched.h which contains task group related information:
struct task_group {
...
...
struct sched_entity **se;
struct cfs_rq **cfs_rq;
...
...
}
The root task group is the task group which belongs every task in system. As we allocated space for the root task group
scheduler entity and runqueue, we go over all possible CPUs ( cpu_possible_mask bitmap) and allocate zeroed memory
from a particular memory node with the kzalloc_node function for the load_balance_mask percpu variable:
DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
Scheduler initialization
137
Linux Inside
Here cpumask_var_t is the cpumask_t with one difference: cpumask_var_t is allocated only nr_cpu_ids bits when the
cpumask_t always has NR_CPUS bits (more about cpumask you can read in the CPU masks part). As you can see:
#ifdef CONFIG_CPUMASK_OFFSTACK
for_each_possible_cpu(i) {
per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
cpumask_size(), GFP_KERNEL, cpu_to_node(i));
}
#endif
this code depends on the CONFIG_CPUMASK_OFFSTACK configuration option. This configuration options says to use dynamic
allocation for cpumask , instead of putting it on the stack. All groups have to be able to rely on the amount of CPU time. With
the call of the two following functions:
init_rt_bandwidth(&def_rt_bandwidth,
global_rt_period(), global_rt_runtime());
init_dl_bandwidth(&def_dl_bandwidth,
global_rt_period(), global_rt_runtime());
we initialize bandwidth management for the SCHED_DEADLINE real-time tasks. These functions initializes rt_bandwidth and
dl_bandwidth structures which are store information about maximum deadline bandwith of the system. For example, let's
As period and runtime we pass result of the global_rt_period and global_rt_runtime functions. Which are 1s second
and and 0.95s by default. The rt_bandwidth structure defined in the kernel/sched/sched.h and looks:
struct rt_bandwidth {
raw_spinlock_t rt_runtime_lock;
ktime_t rt_period;
u64 rt_runtime;
struct hrtimer rt_period_timer;
};
As you can see, it contains runtime and period and also two following fields:
rt_runtime_lock - spinlock for the rt_time protection;
rt_period_timer - high-resolution kernel timer for unthrottled of real-time tasks.
Scheduler initialization
138
Linux Inside
So, in the init_rt_bandwidth we initialize rt_bandwidth period and runtime with the given parameters, initialize the spinlock
and high-resolution time. In the next step, depends on the enabled SMP, we make initialization of the root domain:
#ifdef CONFIG_SMP
init_defrootdomain();
#endif
The real-time scheduler requires global resources to make scheduling decision. But unfortenatelly scalability bottlenecks
appear as the number of CPUs increase. The concept of root domains was introduced for improving scalability. The linux
kernel provides special mechanism for assigning a set of CPUs and memory nodes to a set of task and it is called cpuset . If a cpuset contains non-overlapping with other cpuset CPUs, it is exclusive cpuset . Each exclusive cpuset
defines an isolated domain or root domain of CPUs partitioned from other cpusets or CPUs. A root domain presented by
the struct root_domain from the kernel/sched/sched.h in the linux kernel and its main purpose is to narrow the scope of the
global variables to per-domain variables and all real-time scheduling decisions are made only within the scope of a root
domain. That's all about it, but we will see more details about it in the chapter about scheduling about real-time scheduler.
After root domain initialization, we make initialization of the bandwidth for the real-time tasks of the root task group as we
did it above:
#ifdef CONFIG_RT_GROUP_SCHED
init_rt_bandwidth(&root_task_group.rt_bandwidth,
global_rt_period(), global_rt_runtime());
#endif
In the next step, depends on the CONFIG_CGROUP_SCHED kernel configuration option we initialze the siblings and children
lists of the root task group. As we can read from the documentation, the CONFIG_CGROUP_SCHED is:
This option allows you to create arbitrary task groups using the "cgroup" pseudo
filesystem and control the cpu bandwidth allocated to each such task group.
As we finished with the lists initialization, we can see the call of the autogroup_init function:
#ifdef CONFIG_CGROUP_SCHED
list_add(&root_task_group.list, &task_groups);
INIT_LIST_HEAD(&root_task_group.children);
INIT_LIST_HEAD(&root_task_group.siblings);
autogroup_init(&init_task);
#endif
possible cpu:
for_each_possible_cpu(i) {
struct rq *rq;
...
...
...
Each processor has its own locking and individual runqueue. All runnalble tasks are stored in an active array and indexed
according to its priority. When a process consumes its time slice, it is moved to an expired array. All of these arras are
Scheduler initialization
139
Linux Inside
stored in the special structure which names is runqueu . As there are no global lock and runqueu, we are going through the
all possible CPUs and initialize runqueue for the every cpu. The runque is presented by the rq structure in the linux kernel
which defined in the kernel/sched/sched.h.
rq = cpu_rq(i);
raw_spin_lock_init(&rq->lock);
rq->nr_running = 0;
rq->calc_load_active = 0;
rq->calc_load_update = jiffies + LOAD_FREQ;
init_cfs_rq(&rq->cfs);
init_rt_rq(&rq->rt);
init_dl_rq(&rq->dl);
rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
Here we get the runque for the every CPU with the cpu_rq macto which returns runqueues percpu variable and start to
initialize it with runqueu lock, number of running tasks, calc_load relative fields ( calc_load_active and calc_load_update )
which are used in the reckoning of a CPU load and initialization of the completely fair, real-time and deadline related fields
in a runqueue. After this we initialize cpu_load array with zeros and set the last load update tick to the jiffies variable
which determines the number of time ticks (cycles), since the system boot:
where cpu_load keeps history of runqueue loads in the past, for now CPU_LOAD_IDX_MAX is 5. In the next step we fill
runqueue fields which are related to the SMP, but we will not cover they in this part. And in the end of the loop we initialize
high-resolution timer for the give runqueue and set the iowait (more about it in the separate part about scheduler) number:
init_rq_hrtick(rq);
atomic_set(&rq->nr_iowait, 0);
Now we came out from the for_each_possible_cpu loop and the next we need to set load weight for the init task with the
set_load_weight function. Weight of process is calculated through its dynamic priority which is static priority + scheduling
class of the process. After this we increase memory usage counter of the memory descriptor of the init process and set
scheduler class for the current process:
atomic_inc(&init_mm.mm_count);
current->sched_class = &fair_sched_class;
And make current process (it will be the first init process) idle and update the value of the calc_load_update with the 5
seconds interval:
init_idle(current, smp_processor_id());
calc_load_update = jiffies + LOAD_FREQ;
So, the init process will be run, when there will be no other candidates (as it is the first process in the system). In the end
we just set scheduler_running variable:
scheduler_running = 1;
Scheduler initialization
140
Linux Inside
That's all. Linux kernel scheduler is initialized. Of course, we missed many different details and explanations here, because
we need to know and understand how different concepts (like process and process groups, runqueue, rcu and etc...) works
in the linux kernel , but we took a short look on the scheduler initialization process. All other details we will look in the
separate part which will be fully dedicated to the scheduler.
Conclusion
It is the end of the eighth part about the linux kernel initialization process. In this part, we looked on the initialization process
of the scheduler and we will continue in the next part to dive in the linux kernel initialization process and will see
initialization of the RCU and many more.
and other initialization stuff in the next part.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
CPU masks
high-resolution kernel timer
spinlock
Run queue
Linux kernem memory manager
slub
virtual file system
Linux kernel hotplug documentation
IRQ
Global Descriptor Table
Per-CPU variables
SMP
RCU
CFS Scheduler documentation
Real-Time group scheduling
Previous part
Scheduler initialization
141
Linux Inside
preempt_disable
preempt_enable
for preemption disabling and enabling. First of all let's try to understand what is it preempt in the context of an operating
system kernel. In a simple words, preemption is ability of the operating system kernel to preempt current task to run task
with higher priority. Here we need to disable preemption because we will have only one init process for the early boot
time and we no need to stop it before we will call cpu_idle function. The preempt_disable macro defined in the
include/linux/preempt.h and depends on the CONFIG_PREEMPT_COUNT kernel configuration option. This maco implemeted as:
#define preempt_disable() \
do { \
preempt_count_inc(); \
barrier(); \
} while (0)
Let's look on it. First of all we can see one difference between these macro implementations. The preempt_disable with
CONFIG_PREEMPT_COUNT contains the call of the preempt_count_inc . There is special percpu variable which stores the number
DECLARE_PER_CPU(int, __preempt_count);
In the first implementation of the preempt_disable we increment this __preempt_count . There is API for returning value of
the __preempt_count , it is the preempt_count function. As we called preempt_disable , first of all we increment preemption
counter with the preempt_count_inc macro which expands to the:
where preempt_count_add calls the raw_cpu_add_4 macro which adds 1 to the given percpu variable ( __preempt_count ) in
our case (more about precpu variables you can read in the part about Per-CPU variables). Ok, we increased
__preempt_count and th next step we can see the call of the barrier macro in the both macros. The barrier macro inserts
an optimization barrier. In the processors with x86_64 architecture independent memory access operations can be
performed in any order. That's why we need in the oportunity to point compiler and processor on compliance of order. This
mechanism is memory barrier. Let's consider simple example:
RCU initialization
142
Linux Inside
preempt_disable();
foo();
preempt_enable();
preempt_disable();
preempt_enable();
foo();
In this case non-preemptible function foo can be preempted. As we put barrier macro in the preempt_disable and
preempt_enable macros, it prevents the compiler from swapping preempt_count_inc with other statements. More about
if (WARN(!irqs_disabled(),
"Interrupts were enabled *very* early, fixing it\n"))
local_irq_disable();
which check IRQs state, and disabling (with cli instruction for x86_64 ) if they are enabled.
That's all. Preemption is disabled and we can go ahead.
Here we can see the call of the kmem_cache_create . We already called the kmem_cache_init in the init/main.c. This function
create generalized caches again using the kmem_cache_alloc (more about caches we will see in the Linux kernel memory
management chapter). In our case, as we are using kmem_cache_t it will be used the slab allocator and kmem_cache_create
creates it. As you can seee we pass five parameters to the kmem_cache_create :
name of the cache;
size of the object to store in cache;
offset of the first object in the page;
flags;
constructor for the objects.
and it will create kmem_cache for the integer IDs. Integer IDs is commonly used pattern for the to map set of integer IDs to
the set of pointers. We can see usage of the integer IDs for example in the i2c drivers subsystem. For example
drivers/i2c/i2c-core.c which presentes the core of the i2c subsystem defines ID for the i2c adapter with the DEFINE_IDR
macro:
RCU initialization
143
Linux Inside
static DEFINE_IDR(i2c_adapter_idr);
RCU initialization
The next step is RCU initialization with the rcu_init function and it's implementation depends on two kernel configuration
options:
CONFIG_TINY_RCU
CONFIG_TREE_RCU
In the first case rcu_init will be in the kernel/rcu/tiny.c and in the second case it will be defined in the kernel/rcu/tree.c. We
will see the implementation of the tree rcu , but first of all about the RCU in general.
RCU or read-copy update is a scalable high-performance synchronization mechanism implemented in the Linux kernel. On
the early stage the linux kernel provided support and environment for the concurently running applications, but all execution
was serialized in the kernel using a single global lock. In our days linux kernel has no single global lock, but provides
different mechanisms including lock-free data structures, percpu data structures and other. One of these mechanisms is the read-copy update . The RCU technique designed for rarely-modified data structures. The idea of the RCU is simple. For
example we have a rarely-modified data structure. If somebody wants to change this data structure, we make a copy of this
data structure and make all changes in the copy. In the same time all other users of the data structure use old version of it.
Next, we need to choose safe moment when original version of the data structure will have no users and update it with the
modified copy.
Of course this description of the RCU is very simplified. To understand some details about RCU , first of all we need to learn
some terminology. Data readers in the RCU executed in the critical section. Everytime when data reader joins to the critical
section, it calls the rcu_read_lock , and rcu_read_unlock on exit from the critical section. If the thread is not in the critical
section, it will be in state which called - quiescent state . Every moment when every thread was in the quiescent state
called - grace period . If a thread wants to remove element from the data structure, this occurs in two steps. First steps is
removal - atomically removes element from the data structure, but does not release the physical memory. After this thread-
writer announces and waits while it will be finsihed. From this moment, the removed element is available to the threadreaders. After the grace perioud will be finished, the second step of the element removal will be started, it just removes
element from the physical memory.
There a couple implementations of the RCU . Old RCU called classic, the new implemetation called tree RCU. As you
already can undrestand, the CONFIG_TREE_RCU kernel configuration option enables tree RCU . Another is the tiny RCU
which depends on CONFIG_TINY_RCU and CONFIG_SMP=n . We will see more details about the RCU in general in the separate
RCU initialization
144
Linux Inside
chapter about synchronization primitives, but now let's look on the rcu_init implementation from the kernel/rcu/tree.c:
In the beginning of the rcu_init function we define cpu variable and call rcu_bootup_announce . The rcu_bootup_announce
function is pretty simple:
It just prints information about the RCU with the pr_info function and rcu_bootup_announce_oddness which uses pr_info
too, for printing different information about the current RCU configuration which depends on different kernel configuration
options like CONFIG_RCU_TRACE , CONFIG_PROVE_RCU , CONFIG_RCU_FANOUT_EXACT and etc... In the next step, we can see the call of
the rcu_init_geometry function. This function defined in the same source code file and computes the node tree geometry
depends on amount of CPUs. Actually RCU provides scalability with extremely low internal to RCU lock contention. What if
a data structure will be read from the different CPUs? RCU API provides the rcu_state structure wihch presents RCU
global state including node hierarchy. Hierachy presented by the:
array of structures. As we can read in the comment which is above definition of this structure:
The root (first level) of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]), and the third level
in ->node[m+1] and following (->node[m+1] referenced by ->level[2]). The number of levels is
determined by the number of CPUs and by CONFIG_RCU_FANOUT.
Small systems will have a "hierarchy" consisting of a single rcu_node.
The rcu_node structure defined in the kernel/rcu/tree.h and contains information about current grace period, is grace period
completed or not, CPUs or groups that need to switch in order for current grace period to proceed and etc... Every
rcu_node contains a lock for a couple of CPUs. These rcu_node structures embedded into a linear array in the rcu_state
structure and represeted as a tree with the root in the zero element and it covers all CPUs. As you can see the number of
the rcu nodes determined by the NUM_RCU_NODES which depends on number of available CPUs:
RCU initialization
145
Linux Inside
where levels values depend on the CONFIG_RCU_FANOUT_LEAF configuration option. For example for the simplest case, one
rcu_node will cover two CPU on machine with the eight CPUs:
+-----------------------------------------------------------------+
| rcu_state |
| +----------------------+ |
| | root | |
| | rcu_node | |
| +----------------------+ |
| | | |
| +----v-----+ +--v-------+ |
| | | | | |
| | rcu_node | | rcu_node | |
| | | | | |
| +------------------+ +----------------+ |
| | | | | |
| | | | | |
| +----v-----+ +-------v--+ +-v--------+ +-v--------+ |
| | | | | | | | | |
| | rcu_node | | rcu_node | | rcu_node | | rcu_node | |
| | | | | | | | | |
| +----------+ +----------+ +----------+ +----------+ |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
+---------|-----------------|-------------|---------------|-------+
| | | |
+---------v-----------------v-------------v---------------v--------+
| | | | |
| CPU1 | CPU3 | CPU5 | CPU7 |
| | | | |
| CPU2 | CPU4 | CPU6 | CPU8 |
| | | | |
+------------------------------------------------------------------+
So, in the rcu_init_geometry function we just need to calculate the total number of rcu_node structures. We start to do it
with the calculation of the jiffies till to the first and next fqs which is force-quiescent-state (read above about it):
where:
As we calculated these jiffies, we check that previous defined jiffies_till_first_fqs and jiffies_till_next_fqs variables
are equal to the ULONG_MAX (their default values) and set they equal to the calculated value. As we did not touch these
variables before, they are equal to the ULONG_MAX :
RCU initialization
146
Linux Inside
In the next step of the rcu_init_geometry , we check that rcu_fanout_leaf didn't chage (it has the same value as
CONFIG_RCU_FANOUT_LEAF in compile-time) and equal to the value of the CONFIG_RCU_FANOUT_LEAF configuration option, we just
return:
After this we need to compute the number of nodes that can be handled an rcu_node tree with the given number of levels:
rcu_capacity[0] = 1;
rcu_capacity[1] = rcu_fanout_leaf;
for (i = 2; i <= MAX_RCU_LVLS; i++)
rcu_capacity[i] = rcu_capacity[i - 1] * CONFIG_RCU_FANOUT;
And in the last step we calcluate the number of rcu_nodes at each level of the tree in the loop.
As we calculated geometry of the rcu_node tree, we need to back to the rcu_init function and next step we need to
initialize two rcu_state structures with the rcu_init_one function:
rcu_init_one(&rcu_bh_state, &rcu_bh_data);
rcu_init_one(&rcu_sched_state, &rcu_sched_data);
About this states you can read here. As I wrote above we need to initialize rcu_state structures and rcu_init_one function
will help us with it. After the rcu_state initialization, we can see the call of the __rcu_init_preempt which depends on the
CONFIG_PREEMPT_RCU kernel configuration option. It does the same that previous functions - initialization of the
rcu_preempt_state structure with the rcu_init_one function which has rcu_state type. After this, in the rcu_init , we can
open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
function. This function registers a handler of the pending interrupt . Pending interrupt or softirq supposes that part of
actions cab be delayed for later execution when the system will be less loaded. Pending interrupts represeted by the
following structure:
struct softirq_action
{
void (*action)(struct softirq_action *);
};
RCU initialization
147
Linux Inside
which defined in the include/linux/interrupt.h and contains only one field - handler of an interrupt. You can know about
softirqs in the your system with the:
$ cat /proc/softirqs
CPU0 CPU1 CPU2 CPU3 CPU4 CPU5 CPU6 CPU7
HI: 2 0 0 1 0 2 0 0
TIMER: 137779 108110 139573 107647 107408 114972 99653 98665
NET_TX: 1127 0 4 0 1 1 0 0
NET_RX: 334 221 132939 3076 451 361 292 303
BLOCK: 5253 5596 8 779 2016 37442 28 2855
BLOCK_IOPOLL: 0 0 0 0 0 0 0 0
TASKLET: 66 0 2916 113 0 24 26708 0
SCHED: 102350 75950 91705 75356 75323 82627 69279 69914
HRTIMER: 510 302 368 260 219 255 248 246
RCU: 81290 68062 82979 69015 68390 69385 63304 63473
In our case the interrupt handler is - rcu_process_callbacks which defined in the kernel/rcu/tree.c and does the RCU core
processing for the current CPU. After we registered softirq interrupt for the RCU , we can see the following code:
cpu_notifier(rcu_cpu_notify, 0);
pm_notifier(rcu_pm_notify, 0);
for_each_online_cpu(cpu)
rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu);
Here we can see registration of the cpu notifier which needs in sysmtems which supports CPU hotplug and we will not dive
into details about this theme. The last function in the rcu_init is the rcu_early_boot_tests :
void rcu_early_boot_tests(void)
{
pr_info("Running RCU self tests\n");
if (rcu_self_test)
early_boot_test_call_rcu();
if (rcu_self_test_bh)
early_boot_test_call_rcu_bh();
if (rcu_self_test_sched)
early_boot_test_call_rcu_sched();
}
148
Linux Inside
Ok, we already passed the main theme of this part which is RCU initialization, but it is not the end of the linux kernel
initialization process. In the last paragraph of this theme we will see a couple of functions which work in the initialization
time, but we will not dive into deep details around this function by different reasons. Some reasons not to dive into details
are following:
They are not very important for the generic kernel initialization process and can depend on the different kernel
configuration;
They have the character of debugging and not important too for now;
We will see many of this stuff in the separate parts/chapters.
After we initilized RCU , the next step which you can see in the init/main.c is the - trace_init function. As you can
understand from its name, this function initialize tracing subsystem. More about linux kernel trace system you can read here.
After the trace_init , we can see the call of the radix_tree_init . If you are familar with the different data structures, you
can understand from the name of this function that it initializes kernel implementation of the Radix tree. This function
defined in the lib/radix-tree.c and more about it you can read in the part about Radix tree.
In the next step we can see the functions which are related to the interrupts handling subsystem, they are:
early_irq_init
init_IRQ
softirq_init
We will see explanation about this functions and their implementation in the special part about interrupts and exceptions
handling. After this many different functions (like init_timers , hrtimers_init , time_init and etc...) which are related to
different timing and timers stuff. More about these function we will see in the chapter about timers.
The next couple of functions related with the perf events - perf_event-init (will be separate chapter about perf),
initialization of the profiling with the profile_init . After this we enable irq with the call of the:
local_irq_enable();
which expands to the sti instruction and making post initialization of the SLAB with the call of the kmem_cache_init_late
function (As I wrote above we will know about the SLAB in the Linux memory management chapter).
After the post initialization of the SLAB , next point is initialization of the console with the console_init function from the
drivers/tty/tty_io.c.
After the console initialization, we can see the lockdep_info function which prints information about the Lock dependency
validator. After this, we can see the initialization of the dynamic allocation of the debug objects with the
debug_objects_mem_init , kernel memory leack detector initialization with the kmemleak_init , percpu pageset setup with the
setup_per_cpu_pageset , setup of the NUMA policy with the numa_policy_init , setting time for the scheduler with the
sched_clock_init , pidmap initialization with the call of the pidmap_init function for the initial PID namespace, cache
creation with the anon_vma_init for the private virtual memory areas and early initialization of the ACPI with the
acpi_early_init .
This is the end of the ninth part of the linux kernel initialization process and here we saw initialization of the RCU. In the last
paragraph of this part ( Rest of the initialization process ) we went thorugh the many functions but did not dive into
details about their implementations. Do not worry if you do not know anything about these stuff or you know and do not
understand anything about this. As I wrote already many times, we will see details of implementations, but in the other parts
or other chapters.
RCU initialization
149
Linux Inside
Conclusion
It is the end of the ninth part about the linux kernel initialization process. In this part, we looked on the initialization process
of the RCU subsystem. In the next part we will continue to dive into linux kernel initialization process and I hope that we will
finish with the start_kernel function and will go to the rest_init function from the same init/main.c source code file and
will see that start of the first process.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
lock-free data structures
kmemleak
ACPI
IRQs
RCU
RCU documentation
integer ID management
Documentation/memory-barriers.txt
Runtime locking correctness validator
Per-CPU variables
Linux kernel memory management
slab
i2c
Previous part
RCU initialization
150
Linux Inside
#ifdef CONFIG_X86_ESPFIX64
init_espfix_bsp();
#endif
Here we can see the call of the init_espfix_bsp function which depends on the CONFIG_X86_ESPFIX64 kernel configuration
option. As we can understand from the function name, it does something with the stack. This function defined in the
arch/x86/kernel/espfix_64.c and prevents leaking of 31:16 bits of the esp register during returning to 16-bit stack. First of
all we install espfix page upper directory into the kernel page directory in the init_espfix_bs :
pgd_p = &init_level4_pgt[pgd_index(ESPFIX_BASE_ADDR)];
pgd_populate(&init_mm, pgd_p, (pud_t *)espfix_pud_page);
#define PGDIR_SHIFT 39
#define ESPFIX_PGD_ENTRY _AC(-2, UL)
#define ESPFIX_BASE_ADDR (ESPFIX_PGD_ENTRY << PGDIR_SHIFT)
After we've filled page global directory with the espfix pud, the next step is call of the init_espfix_random and
init_espfix_ap functions. The first function returns random locations for the espfix page and the second enables the
espfix the current CPU. After the init_espfix_bsp finished to work, we can see the call of the thread_info_cache_init
function which defined in the kernel/fork.c and allocates cache for the thread_info if its size is less than PAGE_SIZE :
End of initialization
151
Linux Inside
...
#endif
As we already know the PAGE_SIZE is (_AC(1,UL) << PAGE_SHIFT) or 4096 bytes and THREAD_SIZE is (PAGE_SIZE <<
THREAD_SIZE_ORDER) or 16384 bytes for the x86_64 . The next function after the thread_info_cache_init is the cred_init
from the kernel/cred.c. This function just allocates space for the credentials (like uid , gid and etc...):
more about credentials you can read in the Documentation/security/credentials.txt. Next step is the fork_init function from
the kernel/fork.c. The fork_init function allocates space for the task_struct . Let's look on the implementation of the
fork_init . First of all we can see definitions of the ARCH_MIN_TASKALIGN macro and creation of a slab where task_structs will
be allocated:
#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
#ifndef ARCH_MIN_TASKALIGN
#define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
#endif
task_struct_cachep =
kmem_cache_create("task_struct", sizeof(struct task_struct),
ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
#endif
As we can see this code depends on the CONFIG_ARCH_TASK_STRUCT_ACLLOCATOR kernel configuration option. This configuration
option shows the presence of the alloc_task_struct for the given architecture. As x86_64 has no alloc_task_struct
function, this code will not work and even will not be compiled on the x86_64 .
void arch_task_cache_init(void)
{
task_xstate_cachep =
kmem_cache_create("task_xstate", xstate_size,
__alignof__(union thread_xstate),
SLAB_PANIC | SLAB_NOTRACK, NULL);
setup_xstate_comp();
}
The arch_task_cache_init does initialization of the architecture-specific caches. In our case it is x86_64 , so as we can see,
the arch_task_cache_init allocates space for the task_xstate which represents FPU state and sets up offsets and sizes of
all extended states in xsave area with the call of the setup_xstate_comp function. After the arch_task_cache_init we
calculate default maximum number of threads with the:
set_max_threads(MAX_THREADS);
End of initialization
152
Linux Inside
init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
init_task.signal->rlim[RLIMIT_SIGPENDING] =
init_task.signal->rlim[RLIMIT_NPROC];
As we know the init_task is an instance of the task_struct structure, so it contains signal field which represents signal
handler. It has following type struct signal_struct . On the first two lines we can see setting of the current and maximum
limit of the resource limits . Every process has an associated set of resource limits. These limits specify amount of
resources which current process can use. Here rlim is resource control limit and presented by the:
struct rlimit {
__kernel_ulong_t rlim_cur;
__kernel_ulong_t rlim_max;
};
structure from the include/uapi/linux/resource.h. In our case the resource is the RLIMIT_NPROC which is the maximum
number of process that use can own and RLIMIT_SIGPENDING - the maximum number of pending signals. We can see it in
the:
cat /proc/self/limits
Limit Soft Limit Hard Limit Units
...
...
...
Max processes 63815 63815 processes
Max pending signals 63815 63815 signals
...
...
...
mm_cachep = kmem_cache_create("mm_struct",
sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
After this we allocate SLAB cache for the important vm_area_struct which used by the kernel to manage virtual memory
End of initialization
153
Linux Inside
space:
Note, that we use KMEM_CACHE macro here instead of the kmem_cache_create . This macro defined in the include/linux/slab.h
and just expands to the kmem_cache_create call:
The KMEM_CACHE has one difference from kmem_cache_create . Take a look on __alignof__ operator. The KMEM_CACHE macro
aligns SLAB to the size of the given structure, but kmem_cache_create uses given value to align space. After this we can see
the call of the mmap_init and nsproxy_cache_init functions. The first function initalizes virtual memory area SLAB and the
second function initializes SLAB for namespaces.
The next function after the proc_caches_init is buffer_init . This function defined in the fs/buffer.c source code file and
allocate cache for the buffer_head . The buffer_head is a special structure which defined in the include/linux/buffer_head.h
and used for managing buffers. In the start of the bufer_init function we allocate cache for the struct buffer_head
structures with the call of the kmem_cache_create function as we did it in the previous functions. And calcuate the maximum
size of the buffers in memory with:
which will be equal to the 10% of the ZONE_NORMAL (all RAM from the 4GB on the x86_64 ). The next function after the
buffer_init is - vfs_caches_init . This function allocates SLAB caches and hashtable for different VFS caches. We already
saw the vfs_caches_init_early function in the eighth part of the linux kernel initialization process which initialized caches
for dcache (or directory-cache) and inode cache. The vfs_caches_init function makes post-early initialization of the
dcache and inode caches, private data cache, hash tables for the mount points and etc... More details about VFS will be
described in the separate part. After this we can see signals_init function. This function defined in the kernel/signal.c and
allocates a cache for the sigqueue structures which represents queue of the real time signals. The next function is
page_writeback_init . This function initializes the ratio for the dirty pages. Every low-level page entry contains the dirty bit
err = register_filesystem(&proc_fs_type);
if (err)
return;
As I wrote above we will not dive into details about VFS and different filesystems in this chapter, but will see it in the chapter
about the VFS . After we've registered a new filesystem in the our system, we call the proc_self_init function from the
TOfs/proc/self.c and this function allocates inode number for the self ( /proc/self directory refers to the process
accessing the /proc filesystem). The next step after the proc_self_init is proc_setup_thread_self which setups the
/proc/thread-self directory which contains information about current thread. After this we create /proc/self/mounts
End of initialization
154
Linux Inside
symllink which will contains mount points with the call of the
#ifdef CONFIG_SYSVIPC
proc_mkdir("sysvipc", NULL);
#endif
proc_mkdir("fs", NULL);
proc_mkdir("driver", NULL);
proc_mkdir("fs/nfsd", NULL);
#if defined(CONFIG_SUN_OPENPROMFS) || defined(CONFIG_SUN_OPENPROMFS_MODULE)
proc_mkdir("openprom", NULL);
#endif
proc_mkdir("bus", NULL);
...
...
...
if (!proc_mkdir("tty", NULL))
return;
proc_mkdir("tty/ldisc", NULL);
...
...
...
In the end of the proc_root_init we call the proc_sys_init function which creates /proc/sys directory and initializes the
Sysctl.
It is the end of start_kernel function. I did not describe all functions which are called in the start_kernel . I missed it,
because they are not so important for the generic kernel initialization stuff and depend on only different kernel
configurations. They are taskstats_init_early which exports per-task statistic to the user-space, delayacct_init initializes per-task delay accounting, key_init and security_init initialize diferent security stuff, check_bugs - makes fix up
of the some architecture-dependent bugs, ftrace_init function executes initialization of the ftrace, cgroup_init makes
initialization of the rest of the cgroup subsystem and etc... Many of these parts and subsystems will be described in the
other chapters.
That's all. Finally we passed through the long-long start_kernel function. But it is not the end of the linux kernel
initialization process. We haven't run the first process yet. In the end of the start_kernel we can see the last call of the rest_init function. Let's go ahead.
rcu_scheduler_starting();
smpboot_thread_init();
The first rcu_scheduler_starting makes RCU scheduler active and the second smpboot_thread_init registers the
smpboot_thread_notifier CPU notifier (more about it you can read in the CPU hotplug documentation. After this we can see
End of initialization
155
Linux Inside
Here the kernel_thread function (defined in the kernel/fork.c) creates new kernel thread.As we can see the kernel_thread
function takes three arguments:
Function which will be executed in a new thread;
Parameter for the kernel_init function;
Flags.
We will not dive into details about kernel_thread implementation (we will see it in the chapter which will describe scheduler,
just need to say that kernel_thread invokes clone). Now we only need to know that we create new kernel thread with
kernel_thread function, parent and child of the thread will use shared information about a filesystem and it will start to
execute kernel_init function. A kernel thread differs from an user thread that it runs in a kernel mode. So with these two
kernel_thread calls we create two new kernel threads with the PID = 1 for init process and PID = 2 for kthread . We
already know what is init process. Let's look on the kthread . It is special kernel thread which allows to init and
different parts of the kernel to create another kernel threads. We can see it in the output of the ps util:
Let's postpone kernel_init and kthreadd for now and will go ahead in the rest_init . In the next step after we have
created two new kernel threads we can see the following code:
rcu_read_lock();
kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
rcu_read_unlock();
The first rcu_read_lock function marks the beginning of an RCU read-side critical section and the rcu_read_unlock marks
the end of an RCU read-side critical section. We call these functions because we need to protect the find_task_by_pid_ns .
The find_task_by_pid_ns returns pointer to the task_struct by the given pid. So, here we are getting the pointer to the
task_struct for the PID = 2 (we got it after kthreadd creation with the kernel_thread ). In the next step we call complete
function
complete(&kthreadd_done);
#define DECLARE_COMPLETION(work) \
struct completion work = COMPLETION_INITIALIZER(work)
and expands to the definition of the completion structure. This structure defined in the include/linux/completion.h and
presents completions concept. Completions are a code synchronization mechanism which is provide race-free solution for
the threads that must wait for some process to have reached a point or a specific state. Using completions consists of three
parts: The first is definition of the complete structure and we did it with the DECLARE_COMPLETION . The second is call of the
wait_for_completion . After the call of this function, a thread which called it will not continue to execute and will wait while
other thread did not call complete function. Note that we call wait_for_completion with the kthreadd_done in the beginning
End of initialization
156
Linux Inside
of the kernel_init_freeable :
wait_for_completion(&kthreadd_done);
And the last step is to call complete function as we saw it above. After this the kernel_init_freeable function will not be
executed while kthreadd thread will not be set. After the kthreadd was set, we can see three following functions in the
rest_init :
init_idle_bootup_task(current);
schedule_preempt_disabled();
cpu_startup_entry(CPUHP_ONLINE);
The first init_idle_bootup_task function from the kernel/sched/core.c sets the Scheduling class for the current process
( idle class in our case):
where idle class is a low priority tasks and tasks can be run only when the processor doesn't have to run anything
besides this tasks. The second function schedule_preempt_disabled disables preempt in idle tasks. And the third function
cpu_startup_entry defined in the kernel/sched/idle.c and calls cpu_idle_loop from the kernel/sched/idle.c. The
cpu_idle_loop function works as process with PID = 0 and works in the background. Main purpose of the cpu_idle_loop is
usage of the idle CPU cycles. When there are no one process to run, this process starts to work. We have one process with
idle scheduling class (we just set the current task to the idle with the call of the init_idle_bootup_task function), so the
idle thread does not do useful work and checks that there is not active task to switch:
More about it will be in the chapter about scheduler. So for this moment the start_kernel calls the rest_init function
which spawns an init ( kernel_init function) process and become idle process itself. Now is time to look on the
kernel_init . Execution of the kernel_init function starts from the call of the kernel_init_freeable function. The
kernel_init_freeable function first of all waits for the completion of the kthreadd setup. I already wrote about it above:
wait_for_completion(&kthreadd_done);
After this we set gfp_allowed_mask to __GFP_BITS_MASK which means that already system is running, set allowed cpus/mems
to all CPUs and NUMA nodes with the set_mems_allowed function, allow init process to run on any CPU with the
set_cpus_allowed_ptr , set pid for the cad or Ctrl-Alt-Delete , do preparation for booting of the other CPUs with the call of
the smp_prepare_cpus , call early initcalls with the do_pre_smp_initcalls , initialization of the SMP with the smp_init and
End of initialization
157
Linux Inside
initialization of the lockup_detector with the call of the lockup_detector_init and initialize scheduler with the
sched_init_smp .
After this we can see the call of the following functions - do_basic_setup . Before we will call the do_basic_setup function,
our kernel already initialized for this moment. As comment says:
The do_basic_setup will reinitialize cpuset to the active CPUs, initialization of the khelper - which is a kernel thread which
used for making calls out to userspace from within the kernel, initialize tmpfs, initialize drivers subsystem, enable the
user-mode helper workqueue and make post-early call of the initcalls . We can see openinng of the dev/console and dup
twice file descriptors from 0 to 2 after the do_basic_setup :
We are using two system calls here sys_open and sys_dup . In the next chapters we will see explanation and
implementation of the different system calls. After we opened initial console, we check that rdinit= option was passed to
the kernel command line or set default path of the ramdisk:
if (!ramdisk_execute_command)
ramdisk_execute_command = "/init";
Check user's permissions for the ramdisk and call the prepare_namespace function from the init/do_mounts.c which checks
and mounts the initrd:
This is the end of the kernel_init_freeable function and we need return to the kernel_init . The next step after the
kernel_init_freeable finished its execution is the async_synchronize_full . This function waits until all asynchronous
function calls have been done and after it we will call the free_initmem which will release all memory occupied by the
initialization stuff which located between __init_begin and __init_end . After this we protect .rodata with the
mark_rodata_ro and update state of the system from the SYSTEM_BOOTING to the
system_state = SYSTEM_RUNNING;
if (ramdisk_execute_command) {
ret = run_init_process(ramdisk_execute_command);
if (!ret)
return 0;
pr_err("Failed to execute %s (error %d)\n",
ramdisk_execute_command, ret);
}
End of initialization
158
Linux Inside
First of all it checks the ramdisk_execute_command which we set in the kernel_init_freeable function and it will be equal to
the value of the rdinit= kernel command line parameters or /init by default. The run_init_process function fills the first
element of the argv_init array:
which represents arguments of the init program and call do_execve function:
argv_init[0] = init_filename;
return do_execve(getname_kernel(init_filename),
(const char __user *const __user *)argv_init,
(const char __user *const __user *)envp_init);
The do_execve function defined in the include/linux/sched.h and runs program with the given file name and arguments. If
we did not pass rdinit= option to the kernel command line, kernel starts to check the execute_command which is equal to
value of the init= kernel command line parameter:
if (execute_command) {
ret = run_init_process(execute_command);
if (!ret)
return 0;
panic("Requested init %s failed (error %d).",
execute_command, ret);
}
If we did not pass init= kernel command line parameter too, kernel tries to run one of the following executable files:
if (!try_to_run_init_process("/sbin/init") ||
!try_to_run_init_process("/etc/init") ||
!try_to_run_init_process("/bin/init") ||
!try_to_run_init_process("/bin/sh"))
return 0;
panic("No working init found. Try passing init= option to kernel. "
"See Linux Documentation/init.txt for guidance.");
Conclusion
It is the end of the tenth part about the linux kernel initialization process. And it is not only tenth part, but this is the last
part which describes initialization of the linux kernel. As I wrote in the first part of this chapter, we will go through all steps of
the kernel initialization and we did it. We started at the first architecture-independent function - start_kernel and finished
with the launch of the first init process in the our system. I missed details about different subsystem of the kernel, for
example I almost did not cover linux kernel scheduler or we did not see almost anything about interrupts and exceptions
handling and etc... From the next part we will start to dive to the different kernel subsystems. Hope it will be interesting.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
End of initialization
159
Linux Inside
Links
SLAB
xsave
FPU
Documentation/security/credentials.txt
Documentation/x86/x86_64/mm
RCU
VFS
inode
proc
man proc
Sysctl
ftrace
cgroup
CPU hotplug documentation
completions - wait for completion handling
NUMA
cpus/mems
initcalls
Tmpfs
initrd
panic
Previous part
End of initialization
160
Linux Inside
Interrupts
161
Linux Inside
What is an Interrupt?
We have already heard of the word interrupt in several parts of this book. We even saw a couple of examples of interrupt
handlers. In the current chapter we will start from the theory i.e.
What are interrupts ?
What are interrupt handlers ?
We will then continue to dig deeper into the details of interrupts and how the Linux kernel handles them.
So..., First of all what is an interrupt? An interrupt is an event which is emitted by software or hardware when its needs the
CPU's attention. For example, we press a button on the keyboard and what do we expect next? What should the operating
system and computer do after this? To simplify matters assume that each peripheral device has an interrupt line to the
CPU. A device can use it to signal an interrupt to the CPU. However interrupts are not signaled directly to the CPU. In the
old machines there was a PIC which is a chip responsible for sequentially processing multiple interrupt requests from
multiple devices. In the new machines there is an Advanced Programmable Interrupt Controller commonly known as APIC . An APIC consists of two separate devices:
Local APIC
I/O APIC
The first - Local APIC is located on each CPU core. The local APIC is responsible for handling the CPU-specific interrupt
configuration. The local APIC is usually used to manage interrupts from the APIC-timer, thermal sensor and any other such
locally connected I/O devices.
The second - I/O APIC provides multi-processor interrupt management. It is used to distribute external interrupts among
the CPU cores. More about the local and I/O APICs will be covered later in this chapter. As you can understand, interrupts
can occur at any time. When an interrupt occurs, the operating system must handle it immediately. But what does it mean
to handle an interrupt ? When an interrupt occurs, the operating system must ensure the following steps:
The kernel must pause execution of the current process; (preempt current task)
The kernel must search for the handler of the interrupt and transfer control (execute interrupt handler);
After the interrupt handler completes execution, the interrupted process can resume execution;
Of course there are numerous intricacies involved in this procedure of handling interrupts. But the above 3 steps form the
basic skeleton of the procedure.
Addresses of each of the interrupt handlers are maintained in a special location referred to as the - Interrupt Descriptor
Table or IDT . The processor uses an unique number for recognizing the type of interruption or exception. This number is
called - vector number . A vector number is an index in the IDT . There is limited amount of the vector numbers and it can
be from 0 to 255 . You can note the following range-check upon the vector number within the Linux kernel source-code:
Introduction
162
Linux Inside
You can find this check within the Linux kernel source code related to interrupt setup (eg. The set_intr_gate , void
set_system_intr_gate in arch/x86/include/asm/desc.h). First 32 vector numbers from 0 to 31 are reserved by the
processor and used for the processing of architecture-defined exceptions and interrupts. You can find the table with the
description of these vector numbers in the second part of the Linux kernel initialization process - Early interrupt and
exception handling. Vector numbers from 32 to 255 are designated as user-defined interrupts and are not reserved by the
processor. These interrupts are generally assigned to external I/O devices to enable those devices to send interrupts to the
processor.
Now let's talk about the types of interrupts. Broadly speaking, we can split interrupts into 2 major classes:
External or hardware generated interrupts;
Software-generated interrupts.
The first - external interrupts are received through the Local APIC or pins on the processor which are connected to the
Local APIC . The second - software-generated interrupts are caused by an exceptional condition in the processor itself
(sometimes using special architecture-specific instructions). A common example for an exceptional condition is division by
zero . Another example is exiting a program with the syscall instruction.
As mentioned earlier, an interrupt can occur at any time for a reason which the code and CPU have no control over. On the
other hand, exceptions are synchronous with program execution and can be classified into 3 categories:
Faults
Traps
Aborts
A fault is an exception reported before the execution of a "faulty" instruction (which can then be corrected). If corrected, it
allows the interrupted program to be resume.
Next a trap is an exception which is reported immediately following the execution of the trap instruction. Traps also allow
the interrupted program to be continued just as a fault does.
Finally an abort is an exception that does not always report the exact instruction which caused the exception and does not
allow the interrupted program to be resumed.
Also we already know from the previous part that interrupts can be classified as maskable and non-maskable . Maskable
interrupts are interrupts which can be blocked with the two following instructions for x86_64 - sti and cli . We can find
them in the Linux kernel source code:
and
These two instructions modify the IF flag bit within the interrupt register. The sti instruction sets the IF flag and the cli
Introduction
163
Linux Inside
instruction clears this flag. Non-maskable interrupts are always reported. Usually any failure in the hardware is mapped to
such non-maskable interrupts.
If multiple exceptions or interrupts occur at the same time, the processor handles them in order of their predefined priorities.
We can determine the priorities from the highest to the lowest in the following table:
+----------------------------------------------------------------+
| | |
| Priority | Description |
| | |
+--------------+-------------------------------------------------+
| | Hardware Reset and Machine Checks |
| 1 | - RESET |
| | - Machine Check |
+--------------+-------------------------------------------------+
| | Trap on Task Switch |
| 2 | - T flag in TSS is set |
| | |
+--------------+-------------------------------------------------+
| | External Hardware Interventions |
| | - FLUSH |
| 3 | - STOPCLK |
| | - SMI |
| | - INIT |
+--------------+-------------------------------------------------+
| | Traps on the Previous Instruction |
| 4 | - Breakpoints |
| | - Debug Trap Exceptions |
+--------------+-------------------------------------------------+
| 5 | Nonmaskable Interrupts |
+--------------+-------------------------------------------------+
| 6 | Maskable Hardware Interrupts |
+--------------+-------------------------------------------------+
| 7 | Code Breakpoint Fault |
+--------------+-------------------------------------------------+
| 8 | Faults from Fetching Next Instruction |
| | Code-Segment Limit Violation |
| | Code Page Fault |
+--------------+-------------------------------------------------+
| | Faults from Decoding the Next Instruction |
| | Instruction length > 15 bytes |
| 9 | Invalid Opcode |
| | Coprocessor Not Available |
| | |
+--------------+-------------------------------------------------+
| 10 | Faults on Executing an Instruction |
| | Overflow |
| | Bound error |
| | Invalid TSS |
| | Segment Not Present |
| | Stack fault |
| | General Protection |
| | Data Page Fault |
| | Alignment Check |
| | x87 FPU Floating-point exception |
| | SIMD floating-point exception |
| | Virtualization exception |
+--------------+-------------------------------------------------+
Now that we know a little about the various types of interrupts and exceptions, it is time to move on to a more practical part.
We start with the description of the Interrupt Descriptor Table . As mentioned earlier, the IDT stores entry points of the
interrupts and exceptions handlers. The IDT is similar in structure to the Global Descriptor Table which we saw in the
second part of the Kernel booting process. But of course it has some differences. Instead of descriptors , the IDT entries
are called gates . It can contain one of the following gates:
Interrupt gates
Task gates
Trap gates.
Introduction
164
Linux Inside
in the x86 architecture. Only long mode interrupt gates and trap gates can be referenced in the x86_64 . Like the Global
Descriptor Table , the Interrupt Descriptor table is an array of 8-byte gates on x86 and an array of 16-byte gates on
x86_64 . We can remember from the second part of the Kernel booting process, that Global Descriptor Table must contain
NULL descriptor as its first element. Unlike the Global Descriptor Table , the Interrupt Descriptor Table may contain a
gate; it is not mandatory. For example, you may remember that we have loaded the Interrupt Descriptor table with the NULL
gates only in the earlier part while transitioning into protected mode:
/*
* Set up the IDT
*/
static void setup_idt(void)
{
static const struct gdt_ptr null_idt = {0, 0};
asm volatile("lidtl %0" : : "m" (null_idt));
}
from the arch/x86/boot/pm.c. The Interrupt Descriptor table can be located anywhere in the linear address space and the
base address of it must be aligned on an 8-byte boundary on x86 or 16-byte boundary on x86_64 . Base address of the
IDT is stored in the special register - IDTR . There are two instructions on x86 -compatible processors to modify the IDTR
register:
LIDT
SIDT
The first instruction LIDT is used to load the base-address of the IDT i.e. the specified operand into the IDTR . The second
instruction SIDT is used to read and store the contents of the IDTR into the specified operand. The IDTR register is 48-bits
on the x86 and contains following information:
+-----------------------------------+----------------------+
| | |
| Base address of the IDT | Limit of the IDT |
| | |
+-----------------------------------+----------------------+
47 16 15 0
Looking at the implementation of setup_idt , we have prepared a null_idt and loaded it to the IDTR register with the
lidt instruction. Note that null_idt has gdt_ptr type which is defined as:
struct gdt_ptr {
u16 len;
u32 ptr;
} __attribute__((packed));
Here we can see the definition of the structure with the two fields of 2-bytes and 4-bytes each (a total of 48-bits) as we can
see in the diagram. Now let's look at the IDT entries structure. The IDT entries structure is an array of the 16-byte entries
which are called gates in the x86_64 . They have the following structure:
127 96
+-------------------------------------------------------------------------------+
| |
| Reserved |
| |
+-------------------------------------------------------------------------------95 64
+-------------------------------------------------------------------------------+
| |
| Offset 63..32 |
| |
Introduction
165
Linux Inside
+-------------------------------------------------------------------------------+
63 48 47 46 44 42 39 34 32
+-------------------------------------------------------------------------------+
| | | D | | | | | | |
| Offset 31..16 | P | P | 0 |Type |0 0 0 | 0 | 0 | IST |
| | | L | | | | | | |
-------------------------------------------------------------------------------+
31 16 15 0
+-------------------------------------------------------------------------------+
| | |
| Segment Selector | Offset 15..0 |
| | |
+-------------------------------------------------------------------------------+
To form an index into the IDT, the processor scales the exception or interrupt vector by sixteen. The processor handles the
occurrence of exceptions and interrupts just like it handles calls of a procedure when it sees the call instruction. A
processor uses an unique number or vector number of the interrupt or the exception as the index to find the necessary
Interrupt Descriptor Table entry. Now let's take a closer look at an IDT entry.
As we can see, IDT entry on the diagram consists of the following fields:
0-15 bits - offset from the segment selector which is used by the processor as the base address of the entry point of
And the last Type field describes the type of the IDT entry. There are three different kinds of handlers for interrupts:
Interrupt gate
Trap gate
Task gate
The IST or Interrupt Stack Table is a new mechanism in the x86_64 . It is used as an alternative to the the legacy stackswitch mechanism. Previously The x86 architecture provided a mechanism to automatically switch stack frames in
response to an interrupt. The IST is a modified version of the x86 Stack switching mode. This mechanism unconditionally
switches stacks when it is enabled and can be enabled for any interrupt in the IDT entry related with the certain interrupt
(we will soon see it). From this we can understand that IST is not necessary for all interrupts. Some interrupts can continue
to use the legacy stack switching mode. The IST mechanism provides up to seven IST pointers in the Task State
Segment or TSS which is the special structure which contains information about a process. The TSS is used for stack
switching during the execution of an interrupt or exception handler in the Linux kernel. Each pointer is referenced by an
interrupt gate from the IDT .
The Interrupt Descriptor Table represented by the array of the gate_desc structures:
#ifdef CONFIG_X86_64
...
...
...
Introduction
166
Linux Inside
struct gate_struct64 {
u16 offset_low;
u16 segment;
unsigned ist : 3, zero0 : 5, type : 5, dpl : 2, p : 1;
u16 offset_middle;
u32 offset_high;
u32 zero1;
} __attribute__((packed));
Each active thread has a large stack in the Linux kernel for the x86_64 architecture. The stack size is defined as
THREAD_SIZE and is equal to:
#define PAGE_SHIFT 12
#define PAGE_SIZE (_AC(1,UL) << PAGE_SHIFT)
...
...
...
#define THREAD_SIZE_ORDER (2 + KASAN_STACK_ORDER)
#define THREAD_SIZE (PAGE_SIZE << THREAD_SIZE_ORDER)
The PAGE_SIZE is 4096 -bytes and the THREAD_SIZE_ORDER depends on the KASAN_STACK_ORDER . As we can see, the
KASAN_STACK depends on the CONFIG_KASAN kernel configuration parameter and equals to the:
#ifdef CONFIG_KASAN
#define KASAN_STACK_ORDER 1
#else
#define KASAN_STACK_ORDER 0
#endif
KASan is a runtime memory debugger. So... the THREAD_SIZE will be 16384 bytes if CONFIG_KASAN is disabled or 32768 if this
kernel configuration option is enabled. These stacks contain useful data as long as a thread is alive or in a zombie state.
While the thread is in user-space, the kernel stack is empty except for the thread_info structure (details about this
structure are available in the fourth part of the Linux kernel initialization process) at the bottom of the stack. The active or
zombie threads aren't the only threads with their own stack. There also exist specialized stacks that are associated with
each available CPU. These stacks are active when the kernel is executing on that CPU. When the user-space is executing
on the CPU, these stacks do not contain any useful information. Each CPU has a few special per-cpu stacks as well. The
first is the interrupt stack used for the external hardware interrupts. Its size is determined as follows:
or 16384 bytes. The per-cpu interrupt stack represented by the irq_stack_union union in the Linux kernel for x86_64 :
union irq_stack_union {
char irq_stack[IRQ_STACK_SIZE];
struct {
char gs_base[40];
unsigned long stack_canary;
Introduction
167
Linux Inside
};
};
The first irq_stack field is a 16 kilobytes array. Also you can see that irq_stack_union contains structure with the two
fields:
gs_base - The gs register always points to the bottom of the irqstack union. On the x86_64 , the gs register is
shared by per-cpu area and stack canary (more about per-cpu variables you can read in the special part). All per-cpu
symbols are zero based and the gs points to the base of per-cpu area. You already know that segmented memory
model is abolished in the long mode, but we can set base address for the two segment registers - fs and gs with the
Model specific registers and these registers can be still be used as address registers. If you remember the first part of
the Linux kernel initialization process, you can remember that we have set the gs register:
movl $MSR_GS_BASE,%ecx
movl initial_gs(%rip),%eax
movl initial_gs+4(%rip),%edx
wrmsr
GLOBAL(initial_gs)
.quad INIT_PER_CPU_VAR(irq_stack_union)
stack_canary - Stack canary for the interrupt stack is a stack protector to verify that the stack hasn't been
overwritten. Note that gs_base is an 40 bytes array. GCC requires that stack canary will be on the fixed offset from the
base of the gs and its value must be 40 for the x86_64 and 20 for the x86 .
The irq_stack_union is the first datum in the percpu area, we can see it in the System.map :
0000000000000000 D __per_cpu_start
0000000000000000 D irq_stack_union
0000000000004000 d exception_stacks
0000000000009000 D gdt_page
...
...
...
Now, its time to look at the initialization of the irq_stack_union . Besides the irq_stack_union definition, we can see the
definition of the following per-cpu variables in the arch/x86/include/asm/processor.h:
DECLARE_PER_CPU(char *, irq_stack_ptr);
DECLARE_PER_CPU(unsigned int, irq_count);
The first is the irq_stack_ptr . From the variable's name, it is obvious that this is a pointer to the top of the stack. The
second - irq_count is used to check if a CPU is already on an interrupt stack or not. Initialization of the irq_stack_ptr is
located in the setup_per_cpu_areas function in arch/x86/kernel/setup_percpu.c:
Introduction
168
Linux Inside
Here we go over all the CPUs on-by-one and setup irq_stack_ptr . This turns out to be equal to the top of the interrupt
stack minus 64 . Why 64 ? If you remember, we set the stack canary in the beginning of the start_kernel function from the
init/main.c with the call of the boot_init_stack_canary function:
Note that canary is 64 bits value. That's why we need to subtract 64 from the size of the interrupt stack to avoid
overlapping with the stack canary value. Initialization of the irq_stack_union.gs_base is in the load_percpu_segment function
from the arch/x86/kernel/cpu/common.c:
TODO maybe more about the wrmsl
and as we already know gs register points to the bottom of the interrupt stack:
movl $MSR_GS_BASE,%ecx
movl initial_gs(%rip),%eax
movl initial_gs+4(%rip),%edx
Introduction
169
Linux Inside
wrmsr
GLOBAL(initial_gs)
.quad INIT_PER_CPU_VAR(irq_stack_union)
Here we can see the wrmsr instruction which loads the data from edx:eax into the Model specific register pointed by the
ecx register. In our case model specific register is MSR_GS_BASE which contains the base address of the memory segment
pointed by the gs register. edx:eax points to the address of the initial_gs which is the base address of our
irq_stack_union .
We already know that x86_64 has a feature called Interrupt Stack Table or IST and this feature provides the ability to
switch to a new stack for events non-maskable interrupt, double fault and etc... There can be up to seven IST entries percpu. Some of them are:
DOUBLEFAULT_STACK
NMI_STACK
DEBUG_STACK
MCE_STACK
or
#define DOUBLEFAULT_STACK 1
#define NMI_STACK 2
#define DEBUG_STACK 3
#define MCE_STACK 4
All interrupt-gate descriptors which switch to a new stack with the IST are initialized with the set_intr_gate_ist function.
For example:
where &nmi and &double_fault are addresses of the entries to the given interrupt handlers:
When an interrupt or an exception occurs, the new ss selector is forced to NULL and the ss selectors rpl field is set to
the new cpl . The old ss , rsp , register flags, cs , rip are pushed onto the new stack. In 64-bit mode, the size of
interrupt stack-frame pushes is fixed at 8-bytes, so we will get the following stack:
Introduction
170
Linux Inside
+---------------+
| |
| SS | 40
| RSP | 32
| RFLAGS | 24
| CS | 16
| RIP | 8
| Error code | 0
| |
+---------------+
If the IST field in the interrupt gate is not 0 , we read the IST pointer into rsp . If the interrupt vector number has an error
code associated with it, we then push the error code onto the stack. If the interrupt vector number has no error code, we go
ahead and push the dummy error code on to the stack. We need to do this to ensure stack consistency. Next we load the
segment-selector field from the gate descriptor into the CS register and must verify that the target code-segment is a 64-bit
mode code segment by the checking bit 21 i.e. the L bit in the Global Descriptor Table . Finally we load the offset field
from the gate descriptor into rip which will be the entry-point of the interrupt handler. After this the interrupt handler begins
to execute. After an interrupt handler finishes its execution, it must return control to the interrupted process with the iret
instruction. The iret instruction unconditionally pops the stack pointer ( ss:rsp ) to restore the stack of the interrupted
process and does not depend on the cpl change.
That's all.
Conclusion
It is the end of the first part about interrupts and interrupt handling in the Linux kernel. We saw some theory and the first
steps of the initialization of stuff related to interrupts and exceptions. In the next part we will continue to dive into interrupts
and interrupts handling - into the more practical aspects of it.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me a PR to linux-internals.
Links
Advanced Programmable Interrupt Controller
protected mode
long mode
kernel stacks
PIC
Advanced Programmable Interrupt Controller
long mode
protected mode
Task State Segement
segmented memory model
Model specific registers
Stack canary
Previous chapter
Introduction
171
Linux Inside
void go_to_protected_mode(void)
{
...
setup_idt();
...
}
The setup_idt function defined in the same source code file as the go_to_protected_mode function and just loads address
of the NULL interrupts descriptor table:
where gdt_ptr represents special 48-bit GTDR register which must contain base address of the Global Descriptor Table :
struct gdt_ptr {
u16 len;
u32 ptr;
} __attribute__((packed));
Of course in our case the gdt_ptr does not represent GDTR register, but IDTR since we set Interrupt Descriptor Table .
You will not find idt_ptr structure, because if it had been in the Linux kernel source code, it would have been the same as
gdt_ptr but with different name. So, as you can understand there is no sense to have two similar structures which are
differ only in a name. You can note here, that we do not fill the Interrupt Descriptor Table with entries, because it is too
early to handle any interrupts or exceptions for this moment. That's why we just fill the IDT with the NULL .
And after the setup of the Interrupt descriptor table, Global Descriptor Table and other stuff we jump into protected mode in
the - arch/x86/boot/pmjump.S. More about it you can read in the part which describes transition to the protected mode.
We already know from the earliest parts that entry of the protected mode located in the boot_params.hdr.code32_start and
Start to dive into interrupts
172
Linux Inside
you can see that we pass the entry of the protected mode and boot_params to the protected_mode_jump in the end of the
arch/x86/boot/pm.c:
protected_mode_jump(boot_params.hdr.code32_start,
(u32)&boot_params + (ds() << 4));
The protected_mode_jump defined in the arch/x86/boot/pmjump.S and gets these two parameters in the ax and dx
registers using one of the 8086 calling convention:
GLOBAL(protected_mode_jump)
...
...
...
.byte 0x66, 0xea # ljmpl opcode
2: .long in_pm32 # offset
.word __BOOT_CS # segment
...
...
...
ENDPROC(protected_mode_jump)
GLOBAL(in_pm32)
...
...
jmpl *%eax // %eax contains address of the `startup_32`
...
...
ENDPROC(in_pm32)
As you can remember 32-bit entry point is in the arch/x86/boot/compressed/head_64.S assembly file, although it contains
_64 in the its name. We can see the two similar files in the arch/x86/boot/compressed directory:
arch/x86/boot/compressed/head_32.S .
arch/x86/boot/compressed/head_64.S ;
But the 32-bit mode entry point the the second file in our case. The first file even not compiled for x86_64 . Let's look on the
arch/x86/boot/compressed/Makefile:
We can see here that head_* depends on the $(BITS) variable which depends on the architecture. You can find it in the
arch/x86/Makefile:
ifeq ($(CONFIG_X86_32),y)
...
BITS := 32
else
BITS := 64
...
endif
Now as we jumped on the startup_32 from the arch/x86/boot/compressed/head_64.S we will not find anything related to
Start to dive into interrupts
173
Linux Inside
the interrupt handling here. The startup_32 contains code that makes preparations before transition into the long mode
and directly jumps in it. The long mode entry located startup_64 and it makes preparation before the kernel decompression
that occurs in the decompress_kernel from the arch/x86/boot/compressed/misc.c. After kernel decompressed, we jump on
the startup_64 from the arch/x86/kernel/head_64.S. In the startup_64 we start to build identity-mapped pages. After we
have built identity-mapped pages, checked NX bit, made setup of the Extended Feature Enable Register (see in links),
updated early Global Descriptor Table wit the lgdt instruction, we need to setup gs register with the following code:
movl $MSR_GS_BASE,%ecx
movl initial_gs(%rip),%eax
movl initial_gs+4(%rip),%edx
wrmsr
We already saw this code in the previous part and not time to know better what is going on here. First of all pay attention on
the last wrmsr instruction. This instruction writes data from the edx:eax registers to the model specific register specified by
the ecx register. We can see that ecx contains $MSR_GS_BASE which declared in the arch/x86/include/uapi/asm/msr-index.h
and looks:
From this we can understand that MSR_GS_BASE defines number of the model specific register . Since registers cs , ds ,
es , and ss are not used in the 64-bit mode, their fields are ignored. But we can access memory over fs and gs
registers. The model specific register provides back door to the hidden parts of these segment registers and allows to use
64-bit base address for segment register addressed by the fs and gs . So the MSR_GS_BASE is the hidden part and this part
is mapped on the GS.base field. Let's look on the initial_gs :
GLOBAL(initial_gs)
.quad INIT_PER_CPU_VAR(irq_stack_union)
We pass irq_stack_union symbol to the INIT_PER_CPU_VAR macro which just concatenates init_per_cpu__ prefix with the
given symbol. In our case we will get init_per_cpu__irq_stack_union symbol. Let's look on the linker script. There we can
see following definition:
It tells us that address of the init_per_cpu__irq_stack_union will be irq_stack_union + __per_cpu_load . Now we need to
understand where are init_per_cpu__irq_stack_union and __per_cpu_load and what they mean. The first irq_stack_union
defined in the arch/x86/include/asm/processor.h with the DECLARE_INIT_PER_CPU macro which expands to call of the
init_per_cpu_var macro:
DECLARE_INIT_PER_CPU(irq_stack_union);
#define DECLARE_INIT_PER_CPU(var) \
extern typeof(per_cpu_var(var)) init_per_cpu_var(var)
#define init_per_cpu_var(var) init_per_cpu__##var
If we will expand all macro we will get the same init_per_cpu__irq_stack_union as we got after expanding of the
INIT_PER_CPU macro, but you can note that it is already not just symbol, but variable. Let's look on the
typeof(percpu_var(var)) expression. Our var is irq_stack_union and per_cpu_var macro defined in the
arch/x86/include/asm/percpu.h:
Start to dive into interrupts
174
Linux Inside
where:
#ifdef CONFIG_X86_64
#define __percpu_seg gs
endif
So, we accessing gs:irq_stack_union and geting its type which is irq_union . Ok, we defined the first variable and know its
address, now let's look on the second __per_cpu_load symbol. There are a couple of percpu variables which are located
after this symbol. The __per_cpu_load defined in the include/asm-generic/sections.h:
and presented base address of the per-cpu variables from the data area. So, we know address of the irq_stack_union ,
__per_cpu_load and we know that init_per_cpu__irq_stack_union must be placed right after __per_cpu_load . And we can
...
...
...
ffffffff819ed000 D __init_begin
ffffffff819ed000 D __per_cpu_load
ffffffff819ed000 A init_per_cpu__irq_stack_union
...
...
...
movl $MSR_GS_BASE,%ecx
movl initial_gs(%rip),%eax
movl initial_gs+4(%rip),%edx
wrmsr
Here we specified model specific register with MSR_GS_BASE , put 64-bit address of the initial_gs to the edx:eax pair and
execute wrmsr instruction for filling the gs register with base address of the init_per_cpu__irq_stack_union which will be
bottom of the interrupt stack. After this we will jump to the C code on the x86_64_start_kernel from the
arch/x86/kernel/head64.c. In the x86_64_start_kernel function we do the last preparations before we jump into the generic
and architecture-independent kernel code and on of these preparations is filling of the early Interrupt Descriptor Table
with the interrupts handlres entries or early_idt_handlers . You can remember it, if you have read the part about the Early
interrupt and exception handling and can remember following code:
but I wrote Early interrupt and exception handling part when Linux kernel version was - 3.18 . For this day actual version
of the Linux kernel is 4.1.0-rc6+ and Andy Lutomirski sent the patch and soon it will be in the mainline kernel that
changes behaviour for the early_idt_handlers . NOTE While I wrote this part the patch already turned in the Linux kernel
source code. Let's look on it. Now the same part looks like:
Start to dive into interrupts
175
Linux Inside
AS you can see it has only one difference in the name of the array of the interrupts handlers entry points. Now it is
early_idt_handler_arry :
#define NUM_EXCEPTION_VECTORS 32
#define EARLY_IDT_HANDLER_SIZE 9
So, the early_idt_handler_array is an array of the interrupts handlers entry points and contains one entry point on every
nine bytes. You can remember that previous early_idt_handlers was defined in the arch/x86/kernel/head_64.S. The
early_idt_handler_array is defined in the same source code file too:
ENTRY(early_idt_handler_array)
...
...
...
ENDPROC(early_idt_handler_common)
It fills early_idt_handler_arry with the .rept NUM_EXCEPTION_VECTORS and contains entry of the early_make_pgtable interrupt
handler (more about its implementation you can read in the part about Early interrupt and exception handling). For now we
come to the end of the x86_64 architecture-specific code and the next part is the generic kernel code. Of course you
already can know that we will return to the architecture-specific code in the setup_arch function and other places, but this is
the end of the x86_64 early code.
#ifdef CONFIG_CC_STACKPROTECTOR
...
...
...
#else
static inline void boot_init_stack_canary(void)
{
}
#endif
176
Linux Inside
If the CONFIG_CC_STACKPROTECTOR kernel configuration option is set, the boot_init_stack_canary function starts from the check
stat irq_stack_union that represents per-cpu interrupt stack has offset equal to forty bytes from the stack_canary value:
#ifdef CONFIG_X86_64
BUILD_BUG_ON(offsetof(union irq_stack_union, stack_canary) != 40);
#endif
As we can read in the previous part the irq_stack_union represented by the following union:
union irq_stack_union {
char irq_stack[IRQ_STACK_SIZE];
struct {
char gs_base[40];
unsigned long stack_canary;
};
};
which defined in the arch/x86/include/asm/processor.h. We know that unioun in the C programming language is a data
structure which stores only one field in a memory. We can see here that structure has first field - gs_base which is 40 bytes
size and represents bottom of the irq_stack . So, after this our check with the BUILD_BUG_ON macro should end
successfully. (you can read the first part about Linux kernel initialization process if you're interesting about the BUILD_BUG_ON
macro).
After this we calculate new canary value based on the random number and Time Stamp Counter:
get_random_bytes(&canary, sizeof(canary));
tsc = __native_read_tsc();
canary += tsc + (tsc << 32UL);
and write canary value to the irq_stack_union with the this_cpu_write macro:
this_cpu_write(irq_stack_union.stack_canary, canary);
more about this_cpu_* operation you can read in the Linux kernel documentation.
#ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
...
#define local_irq_disable() \
do { raw_local_irq_disable(); trace_hardirqs_off(); } while (0)
...
#else
...
177
Linux Inside
They are both similar and as you can see have only one difference: the local_irq_disable macro contains call of the
trace_hardirqs_off when CONFIG_TRACE_IRQFLAGS_SUPPORT is enabled. There is special feature in the lockdep subsystem irq-flags tracing for tracing hardirq and stoftirq state. In ourcase lockdep subsytem can give us interesting
information about hard/soft irqs on/off events which are occurs in the system. The trace_hardirqs_off function defined in
the kernel/locking/lockdep.c:
void trace_hardirqs_off(void)
{
trace_hardirqs_off_caller(CALLER_ADDR0);
}
EXPORT_SYMBOL(trace_hardirqs_off);
and just calls trace_hardirqs_off_caller function. The trace_hardirqs_off_caller checks the hardirqs_enabled filed of the
current process increment the redundant_hardirqs_off if call of the local_irq_disable was redundant or the
hardirqs_off_events if it was not. These two fields and other lockdep statistic related fields are defined in the
struct lockdep_stats {
...
...
...
int softirqs_off_events;
int redundant_softirqs_off;
...
...
...
}
If you will set CONFIG_DEBUG_LOCKDEP kernel configuration option, the lockdep_stats_debug_show function will write all tracing
information to the /proc/lockdep :
178
Linux Inside
Ok, now we know a little about tracing, but more info will be in the separate part about lockdep and tracing . You can see
that the both local_disable_irq macros have the same part - raw_local_irq_disable . This macro defined in the
arch/x86/include/asm/irqflags.h and expands to the call of the:
And you already must remember that cli instruction clears the IF flag which determines ability of a processor to handle
and interrupt or an exception. Besides the local_irq_disable , as you already can know there is an inverse macr local_irq_enable . This macro has the same tracing mechanism and very similar on the local_irq_enable , but as you can
understand from its name, it enables interrupts with the sti instruction:
Now we know how local_irq_disable and local_irq_enable work. It was the first call of the local_irq_disable macro, but
we will meet these macros many times in the Linux kernel source code. But for now we are in the start_kernel function
from the init/main.c and we just disabled local interrupts. Why local and why we did it? Previously kernel provided a
method to disable interrupts on all processors and it was called cli . This function was removed and now we have
local_irq_{enabled,disable} to disable or enable interrupts on the current processor. After we've disabled the interrupts
early_boot_irqs_disabled = true;
and used in the different places. For example it used in the smp_call_function_many function from the kernel/smp.c for the
checking possible deadlock when interrupts are disabled:
179
Linux Inside
All of these functions defined in the arch/x86/include/asm/desc.h and do the similar thing but not the same. The first
set_intr_gate_ist function inserts new an interrupt gate in the IDT . Let's look on its implementation:
First of all we can see the check that n which is vector number of the interrupt is not greater than 0xff or 255. We need to
check it because we remember from the previous part that vector number of an interrupt must be between 0 and 255 . In
the next step we can see the call of the _set_gate function that sets a given interrupt gate to the IDT table:
Here we start from the pack_gate function which takes clean IDT entry represented by the gate_desc structure and fills it
with the base address and limit, Interrupt Stack Table, Privilege level, type of an interrupt which can be one of the following
values:
GATE_INTERRUPT
GATE_TRAP
GATE_CALL
GATE_TASK
and set the present bit for the given IDT entry:
static inline void pack_gate(gate_desc *gate, unsigned type, unsigned long func,
unsigned dpl, unsigned ist, unsigned seg)
{
gate->offset_low = PTR_LOW(func);
gate->segment = __KERNEL_CS;
gate->ist = ist;
gate->p = 1;
gate->dpl = dpl;
gate->zero0 = 0;
180
Linux Inside
gate->zero1 = 0;
gate->type = type;
gate->offset_middle = PTR_MIDDLE(func);
gate->offset_high = PTR_HIGH(func);
}
After this we write just filled interrupt gate to the IDT with the write_idt_entry macro which expands to the
native_write_idt_entry and just copy the interrupt gate to the idt_table table by the given index:
That's all. The second set_system_intr_gate_ist function has only one difference from the set_intr_gate_ist :
Do you see it? Look on the fourth parameter of the _set_gate . It is 0x3 . In the set_intr_gate it was 0x0 . We know that
this parameter represent DPL or privilege level. We also know that 0 is the highest privilge level and 3 is the lowest.Now
we know how set_system_intr_gate_ist , set_intr_gate_ist , set_intr_gate are work and we can return to the
early_trap_init function. Let's look on it again:
We set two IDT entries for the #DB interrupt and int3 . These functions takes the same set of parameters:
vector number of an interrupt;
address of an interrupt handler;
interrupt stack table index.
That's all. More about interrupts and handlers you will know in the next parts.
Conclusion
It is the end of the second part about interrupts and interrupt handling in the Linux kernel. We saw the some theory in the
previous part and started to dive into interrupts and exceptions handling in the current part. We have started from the
earliest parts in the Linux kernel source code which are related to the interrupts. In the next part we will continue to dive into
this interesting theme and will know more about interrupt handling process.
If you will have any questions or suggestions write me a comment or ping me at twitter.
181
Linux Inside
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
IDT
Protected mode
List of x86 calling conventions
8086
Long mode
NX
Extended Feature Enable Register
Model-specific register
Process identifier
lockdep
irqflags tracing
IF
Stack canary
Union type
thiscpu* operations
vector number
Interrupt Stack Table
Privilege level
Previous part
182
Linux Inside
exceptions:
#DB - debug exception, transfers control from the interrupted process to the debug handler;
#BP - breakpoint exception, caused by the int 3 instruction.
These exceptions allow the x86_64 architecture to have early exception processing for the purpose of debugging via the
kgdb.
As you can remember we set these exceptions handlers in the early_trap_init function:
from the arch/x86/kernel/traps.c. We already saw implementation of the set_intr_gate_ist and set_system_intr_gate_ist
functions in the previous part and now we will look on the implementation of these early exceptions handlers.
number of the #DB exceptions is 1 (we pass it as X86_TRAP_DB ) and has no error code:
The second is #BP or breakpoint exception occurs when processor executes the INT 3 instruction. We can add it anywhere
in our code, for example let's look on the simple program:
// breakpoint.c
#include <stdio.h>
int main() {
int i;
while (i < 6){
Interrupt handlers
183
Linux Inside
If we will compile and run this program, we will see following output:
But if will run it with gdb, we will see our breakpoint and can continue execution of our program:
$ gdb breakpoint
...
...
...
(gdb) run
Starting program: /home/alex/breakpoints
i equal to: 0
Program received signal SIGTRAP, Trace/breakpoint trap.
0x0000000000400585 in main ()
=> 0x0000000000400585 <main+31>: 83 45 fc 01 add DWORD PTR [rbp-0x4],0x1
(gdb) c
Continuing.
i equal to: 1
Program received signal SIGTRAP, Trace/breakpoint trap.
0x0000000000400585 in main ()
=> 0x0000000000400585 <main+31>: 83 45 fc 01 add DWORD PTR [rbp-0x4],0x1
(gdb) c
Continuing.
i equal to: 2
Program received signal SIGTRAP, Trace/breakpoint trap.
0x0000000000400585 in main ()
=> 0x0000000000400585 <main+31>: 83 45 fc 01 add DWORD PTR [rbp-0x4],0x1
...
...
...
Now we know a little about these two exceptions and we can move on to consideration of their handlers.
You will not find these functions in the C code. All that can be found in in the *.c/*.h files only definition of this functions in
the arch/x86/include/asm/traps.h:
But we can see asmlinkage descriptor here. The asmlinkage is the special specificator of the gcc. Actually for a C
Interrupt handlers
184
Linux Inside
functions which are will be called from assembly, we need in explicit declaration of the function calling convention. In our
case, if function maked with asmlinkage descriptor, then gcc will compile the function to retrieve parameters from stack.
So, both handlers are defined in the arch/x86/kernel/entry_64.S assembly source code file with the idtentry macro:
Actually debug and int3 are not interrupts handlers. Remember that before we can execute an interrupt/exception
handler, we need to do some preparations as:
When an interrupt or exception occured, the processor uses an exception or interrupt vector as an index to a descriptor
in the IDT ;
In legacy mode ss:esp registers are pushed on the stack only if privilege level changed. In 64-bit mode ss:rsp
pushed on the stack everytime;
During stack switching with IST the new ss selector is forced to null. Old ss and rsp are pushed on the new stack.
The rflags , cs , rip and error code pushed on the stack;
Control transfered to an interrupt handler;
After an interrupt handler will finish its work and finishes with the iret instruction, old ss will be poped from the stack
and loaded to the ss register.
ss:rsp will be popped from the stack unconditionally in the 64-bit mode and will be popped only if there is a privilege
+--------------------+
+40 | ss |
+32 | rsp |
+24 | rflags |
+16 | cs |
+8 | rip |
0 | error code |
+--------------------+
Now we can see on the preparations before a process will transfer control to an interrupt/exception handler from practical
side. As I already wrote above the first thirteen exceptions handlers defined in the arch/x86/kernel/entry_64.S assembly file
with the idtentry macro:
This macro defines an exception entry point and as we can see it takes five arguments:
sym - defines global symbol with the .globl name .
do_sym - an interrupt handler.
has_error_code:req - information about error code, The :req qualifier tells the assembler that the argument is
required;
paranoid - shows us how we need to check current mode;
shift_ist - shows us what's stack to use;
185
Linux Inside
The differences are only in the global name and name of exceptions handlers. Now let's look how idtentry macro
implemented. It starts from the two checks:
First check makes the check that an exceptions uses Interrupt stack table and paranoid is set, in other way it emits the
erorr with the .error directive. The second if clause checks existence of an error code and calls XCPT_FRAME or INTR_FRAME
macros depends on it. These macros just expand to the set of CFI directives which are used by GNU AS to manage call
frames. The CFI directives are used only to generate dwarf2 unwind information for better backtraces and they don't
change any code, so we will not go into detail about it and from this point I will skip all code which is related to these
directives. In the next step we check error code again and push it on the stack if an exception has it with the:
.ifeq \has_error_code
pushq_cfi $-1
.endif
The pushq_cfi macro defined in the arch/x86/include/asm/dwarf2.h and expands to the pushq instruction which pushes
given error code:
Pay attention on the $-1 . We already know that when an exception occrus, the processor pushes ss , rsp , rflags , cs
and rip on the stack:
With the pushq \reg we denote that place before the RIP will contain error code of an exception:
The ORIG_RAX will contain error code of an exception, IRQ number on a hardware interrupt and system call number on
system call entry. In the next step we can see thr ALLOC_PT_GPREGS_ON_STACK macro which allocates space for the 15 general
purpose registers on the stack:
Interrupt handlers
186
Linux Inside
After this we check paranoid and if it is set we check first three CPL bits. We compare it with the 3 and it allows us to
know did we come from userspace or not:
.if \paranoid
.if \paranoid == 1
CFI_REMEMBER_STATE
testl $3, CS(%rsp)
jnz 1f
.endif
call paranoid_entry
.else
call error_entry
.endif
If we came from userspace we jump on the label 1 which starts from the call error_entry instruction. The error_entry
saves all registers in the pt_regs structure which presetens an interrupt/exception stack frame and defined in the
arch/x86/include/uapi/asm/ptrace.h. It saves common and extra registers on the stack with the:
SAVE_C_REGS 8
SAVE_EXTRA_REGS 8
from rdi to r15 and executes swapgs instruction. This instruction provides a method to for the Linux kernel to obtain a
pointer to the kernel data structures and save the user's gsbase . After this we will exit from the error_entry with the ret
instruction. After the error_entry finished to execute, since we came from userspace we need to switch on kernel interrupt
stack:
movq %rsp,%rdi
call sync_regs
We just save all registers to the error_entry in the error_entry , we put address of the pt_regs to the rdi and call
sync_regs function from the arch/x86/kernel/traps.c:
This function switchs off the IST stack if we came from usermode. After this we switch on the stack which we got from the
sync_regs :
movq %rax,%rsp
movq %rsp,%rdi
and put pointer of the pt_regs again in the rdi , and in the last step we call an exception handler:
call \do_sym
Interrupt handlers
187
Linux Inside
So, realy exceptions handlers are do_debug and do_int3 functions. We will see these function in this part, but little later.
First of all let's look on the preparations before a processor will transfer control to an interrupt handler. In another way if
paranoid is set, but it is not 1, we call paranoid_entry which makes almost the same that error_entry , but it checks
ENTRY(paranoid_entry)
SAVE_C_REGS 8
SAVE_EXTRA_REGS 8
...
...
movl $MSR_GS_BASE,%ecx
rdmsr
testl %edx,%edx
js 1f /* negative -> in kernel */
SWAPGS
...
...
ret
END(paranoid_entry)
If edx wll be negative, we are in the kernel mode. As we store all registers on the stack, check that we are in the kernel
mode, we need to setup IST stack if it is set for a given exception, call an exception handler and restore the exception
stack:
.if \shift_ist != -1
subq $EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
.endif
call \do_sym
.if \shift_ist != -1
addq $EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
.endif
The last step when an exception handler will finish it's work all registers will be restored from the stack with the
RESTORE_C_REGS and RESTORE_EXTRA_REGS macros and control will be returned an interrupted task. That's all. Now we know
about preparation before an interrupt/exception handler will start to execute and we can go directly to the implementation of
the handlers.
which tells to compiler that something else uses this function (in our case these functions are called from the assembly
interrupt preparation code). And also they takes two parameters:
pointer to the pt_regs structure which contains registers of the interrupted task;
error code.
First of all let's consider do_debug handler. This function starts from the getting previous state with the ist_enter function
from the arch/x86/kernel/traps.c. We call it because we need to know, did we come to the interrupt handler from the kernel
Interrupt handlers
188
Linux Inside
prev_state = ist_enter(regs);
The ist_enter function returns previous state context state and executes a couple preprartions before we continue to
handle an exception. It starts from the check of the previous mode with the user_mode_vm macro. It takes pt_regs structure
which contains a set of registers of the interrupted task and returns 1 if we came from userspace and 0 if we came from
kernel space. According to the previous mode we execute exception_enter if we are from the userspace or inform RCU if
we are from krenel space:
...
if (user_mode_vm(regs)) {
prev_state = exception_enter();
} else {
rcu_nmi_enter();
prev_state = IN_KERNEL;
}
...
...
...
return prev_state;
After this we load the DR6 debug registers to the dr6 variable with the call of the get_debugreg macro from the
arch/x86/include/asm/debugreg.h:
get_debugreg(dr6, 6);
dr6 &= ~DR6_RESERVED;
The DR6 debug register is debug status register contains information about the reason for stopping the #DB or debug
exception handler. After we loaded its value to the dr6 variable we filter out all reserved bits ( 4:12 bits). In the next step
we check dr6 register and previous state with the following if condition expression:
If dr6 does not show any reasons why we caught this trap we set user_icebp to one which means that user-code wants to
get SIGTRAP signal. In the next step we check was it kmemcheck trap and if yes we go to exit:
After we did all these checks, we clear the dr6 register, clear the DEBUGCTLMSR_BTF flag which provides single-step on
branches debugging, set dr6 register for the current thread and increase debug_stack_usage per-cpu) variable with the:
set_debugreg(0, 6);
clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
tsk->thread.debugreg6 = dr6;
debug_stack_usage_inc();
Interrupt handlers
189
Linux Inside
more about local_irq_enabled and related stuff you can read in the second part about interrupts handling in the Linux
kernel. In the next step we check the previous mode was virtual 8086 and handle the trap:
If we came not from the virtual 8086 mode, we need to check dr6 register and previous mode as we did it above. Here we
check if step mode debugging is enabled and we are not from the user mode, we enabled step mode debugging in the dr6
copy in the current thread, set TIF_SINGLE_STEP falg and re-enable Trap flag for the user mode:
si_code = get_si_code(tsk->thread.debugreg6);
In the end we disabled irqs , decrement value of the debug_stack_usage and exit from the exception handler with the
ist_exit function.
The second exception handler is do_int3 defined in the same source code file - arch/x86/kernel/traps.c. In the do_int3 we
makes almost the same that in the do_debug handler. We get the previous state with the ist_enter , increment and
decrement the debug_stack_usage per-cpu variable, enabled and disable local interrupts. But of course there is one
difference between these two handlers. We need to lock and than sync processor cores during breakpoint patching.
That's all.
Interrupt handlers
190
Linux Inside
Conclusion
It is the end of the third part about interrupts and interrupt handling in the Linux kernel. We saw the initialization of the
Interrupt descriptor table in the previous part with the #DB and #BP gates and started to dive into preparation before control
will be transfered to an exception handler and implementation of some interrupt handlers in this part. In the next part we will
continue to dive into this theme and will go next by the setup_arch function and will try to understand interrupts handling
related stuff.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
Debug registers
Intel 80385
INT 3
gcc
TSS
GNU assembly .error directive
dwarf2
CFI directives
IRQ
system call
swapgs
SIGTRAP
Per-CPU variables
kgdb
ACPI
Previous part
Interrupt handlers
191
Linux Inside
function that called in the setup_arch function which defined in the arch/x86/kernel/setup.c. In this part we will continue to
dive into an interrupts and exceptions handling in the Linux kernel for x86_64 and continue to do it from from the place
where we left off in the last part. First thing which is related to the interrupts and exceptions handling is the setup of the #PF
or page fault handler with the early_trap_pf_init function. Let's start from it.
This macro defined in the arch/x86/include/asm/desc.h. We already saw macros like this in the previous part set_system_intr_gate and set_intr_gate_ist . This macro checks that given vector number is not greater than 255
(maximum vector number) and calls _set_gate function as set_system_intr_gate and set_intr_gate_ist did it:
enum {
...
...
...
192
Linux Inside
...
X86_TRAP_PF, /* 14, Page Fault */
...
...
...
}
When the early_trap_pf_init will be called, the set_intr_gate will be expanded to the call of the _set_gate which will fill
the IDT with the handler for the page fault. Now let's look on the implementation of the page_fault handler. The
page_fault handler defined in the arch/x86/kernel/entry_64.S assembly source code file as all exceptions handlers. Let's
look on it:
We saw in the previous part how #DB and #BP handlers defined. They were defined with the idtentry macro, but here we
can see trace_idtentry . This macro defined in the same source code file and depends on the CONFIG_TRACING kernel
configuration option:
#ifdef CONFIG_TRACING
.macro trace_idtentry sym do_sym has_error_code:req
idtentry trace(\sym) trace(\do_sym) has_error_code=\has_error_code
idtentry \sym \do_sym has_error_code=\has_error_code
.endm
#else
.macro trace_idtentry sym do_sym has_error_code:req
idtentry \sym \do_sym has_error_code=\has_error_code
.endm
#endif
We will not dive into exceptions Tracing now. If CONFIG_TRACING is not set, we can see that trace_idtentry macro just
expands to the normal idtentry . We already saw implementation of the idtentry macro in the previous part, so let's start
from the page_fault exception handler.
As we can see in the idtentry definition, the handler of the page_fault is do_page_fault function which defined in the
arch/x86/mm/fault.c and as all exceptions handlers it takes two arguments:
regs - pt_regs structure that holds state of an interrupted process;
error_code - error code of the page fault exception.
Let's look inside this function. First of all we read content of the cr2 control register:
This register contains a linear address which caused page fault . In the next step we make a call of the exception_enter
function from the include/linux/context_tracking.h. The exception_enter and exception_exit are functions from context
tracking subsytem in the Linux kernel used by the RCU to remove its dependency on the timer tick while a processor runs
in userspace. Almost in the every exception handler we will see similar code:
193
Linux Inside
prev_state = exception_enter();
...
... // exception handler here
...
exception_exit(prev_state);
The exception_enter function checks that context tracking is enabled with the context_tracking_is_enabled and if it is in
enabled state, we get previous context with te this_cpu_read (more about this_cpu_* operations you can read in the
Documentation). After this it calls context_tracking_user_exit function which informs that Inform the context tracking that
the processor is exiting userspace mode and entering the kernel:
enum ctx_state {
IN_KERNEL = 0,
IN_USER,
} state;
And in the end we return previous context. Between the exception_enter and exception_exit we call actual page fault
handler:
The __do_page_fault is defined in the same source code file as do_page_fault - arch/x86/mm/fault.c. In the bingging of the
__do_page_fault we check state of the kmemcheck checker. The kmemcheck detects warns about some uses of uninitialized
if (kmemcheck_active(regs))
kmemcheck_hide(regs);
prefetchw(&mm->mmap_sem);
After this we can see the call of the prefetchw which executes instruction with the same name which fetches
X86_FEATURE_3DNOW to get exclusive cache line. The main purpose of prefetching is to hide the latency of a memory
access. In the next step we check that we got page fault not in the kernel space with the following conditiion:
if (unlikely(fault_in_kernel_space(address))) {
...
...
...
}
194
Linux Inside
or 0x00007ffffffff000 . Pay attention on unlikely macro. There are two macros in the Linux kernel:
You can often find these macros in the code of the Linux kernel. Main purpose of these macros is optimization. Sometimes
this situation is that we need to check the condition of the code and we know that it will rarely be true or false . With
these macros we can tell to the compiler about this. For example
Here we can see proc_root_readdir function which will be called when the Linux VFS needs to read the root directory
contents. If condition marked with unlikely , compiler can put false code right after branching. Now let's back to the our
address check. Comparison between the given address and the 0x00007ffffffff000 will give us to know, was page fault in
the kernel mode or user mode. After this check we know it. After this __do_page_fault routine will try to understand the
problem that provoked page fault exception and then will pass address to the approprite routine. It can be kmemcheck fault,
spurious fault, kprobes fault and etc. Will not dive into implementation details of the page fault exception handler in this
part, because we need to know many different concepts which are provided by the Linux kerne, but will see it in the chapter
about the memory management in the Linux kernel.
Back to start_kernel
There are many different function calls after the early_trap_pf_init in the setup_arch function from different kernel
subsystems, but there are no one interrupts and exceptions handling related. So, we have to go back where we came from
- start_kernel function from the init/main.c. The first things after the setup_arch is the trap_init function from the
arch/x86/kernel/traps.c. This function makes initialization of the remaining exceptions handlers (remember that we already
setup 3 handlres for the #DB - debug exception, #BP - breakpoint exception and #PF - page fault exception). The
trap_init function starts from the check of the Extended Industry Standard Architecture:
#ifdef CONFIG_EISA
void __iomem *p = early_ioremap(0x0FFFD9, 4);
if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
EISA_bus = 1;
early_iounmap(p, 4);
195
Linux Inside
#endif
Note that it depends on the CONFIG_EISA kernel configuration parameter which represetns EISA support. Here we use
early_ioremap function to map I/O memory on the page tables. We use readl function to read first 4 bytes from the
mapped region and if they are equal to EISA string we set EISA_bus to one. In the end we just unmap previously mapped
region. More about early_ioremap you can read in the part which describes Fix-Mapped Addresses and ioremap.
After this we start to fill the Interrupt Descriptor Table with the different interrupt gates. First of all we set #DE or Divide
Error and #NMI or Non-maskable Interrupt :
set_intr_gate(X86_TRAP_DE, divide_error);
set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
We use set_intr_gate macro to set the interrupt gate for the #DE exception and set_intr_gate_ist for the #NMI . You can
remember that we already used these macros when we have set the interrupts gates for the page fault handler, debug
handler and etc, you can find explanation of it in the previous part. After this we setup exception gates for the following
exceptions:
set_system_intr_gate(X86_TRAP_OF, &overflow);
set_intr_gate(X86_TRAP_BR, bounds);
set_intr_gate(X86_TRAP_UD, invalid_op);
set_intr_gate(X86_TRAP_NM, device_not_available);
was executed;
#BR or BOUND Range exceeded exception. This exception indeicates that a BOUND-range-exceed fault occured when a
This exception occurs when processor detected a second exception while calling an exception handler for a prior
exception. In usual way when the processor detects another exception while trying to call an exception handler, the two
exceptions can be handled serially. If the processor cannot handle them serially, it signals the double-fault or #DF
exception.
The following set of the interrupt gates is:
set_intr_gate(X86_TRAP_OLD_MF, &coprocessor_segment_overrun);
set_intr_gate(X86_TRAP_TS, &invalid_TSS);
set_intr_gate(X86_TRAP_NP, &segment_not_present);
set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
set_intr_gate(X86_TRAP_GP, &general_protection);
set_intr_gate(X86_TRAP_SPURIOUS, &spurious_interrupt_bug);
set_intr_gate(X86_TRAP_MF, &coprocessor_error);
set_intr_gate(X86_TRAP_AC, &alignment_check);
196
Linux Inside
called general-protection violations. There are many different conditions that can cause general-procetion exception.
For example loading the ss , ds , es , fs , or gs register with a segment selector for a system segment, writing to a
code segment or a read-only data segment, referencing an entry in the Interrupt Descriptor Table (following an
interrupt or exception) that is not an interrupt, trap, or task gate and many many more.
Spurious Interrupt - a hardware interrupt that is unwanted.
#MF or x87 FPU Floating-Point Error exception caused when the x87 FPU has detected a floating point error.
#AC or Alignment Check exception Indicates that the processor detected an unaligned memory operand when
#ifdef CONFIG_X86_MCE
set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
#endif
Note that it depends on the CONFIG_X86_MCE kernel configuration option and indicates that the processor detected an
internal machine error or a bus error, or that an external agent detected a bus error. The next exception gate is for the
SIMD Floating-Point exception:
set_intr_gate(X86_TRAP_XF, &simd_coprocessor_error);
which indicates the processor has detected an SSE or SSE2 or SSE3 SIMD floating-point exception. There are six classes
of numeric exception conditions that can occur while executing an SIMD floating-point instruction:
Invalid operation
Divide-by-zero
Denormal operand
Numeric overflow
Numeric underflow
Inexact result (Precision)
In the next step we fill the used_vectors array which defined in the arch/x86/include/asm/desc.h header file and represents
bitmap :
DECLARE_BITMAP(used_vectors, NR_VECTORS);
of the first 32 interrupts (more about bitmaps in the Linux kernel you can read in the part which describes cpumasks and
bitmaps)
197
Linux Inside
set_bit(i, used_vectors)
After this we setup the interrupt gate for the ia32_syscall and add 0x80 to the used_vectors bitmap:
#ifdef CONFIG_IA32_EMULATION
set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
set_bit(IA32_SYSCALL_VECTOR, used_vectors);
#endif
There is CONFIG_IA32_EMULATION kernel configuration option on x86_64 Linux kernels. This option provides ability to execute
32-bit processes in compatibility-mode. In the next parts we will see how it works, in the meantime we need only to know
that there is yet another interrupt gate in the IDT with the vector number 0x80 . In the next step we maps IDT to the fixmap
area:
and write its address to the idt_descr.address (more about fix-mapped addresses you can read in the second part of the
Linux kernel memory management chapter). After this we can see the call of the cpu_init function that defined in the
arch/x86/kernel/cpu/common.c. This function makes initialization of the all per-cpu state. In the beginnig of the cpu_init
we do the following things: First of all we wait while current cpu is initialized and than we call the cr4_init_shadow function
which stores shadow copy of the cr4 control register for the current cpu and load CPU microcode if need with the following
function calls:
wait_for_master_cpu(cpu);
cr4_init_shadow();
load_ucode_ap();
Next we get the Task State Segement for the current cpu and orig_ist structure which represents origin Interrupt Stack
Table values with the:
t = &per_cpu(cpu_tss, cpu);
oist = &per_cpu(orig_ist, cpu);
As we got values of the Task State Segement and Interrupt Stack Table for the current processor, we clear following bits in
the cr4 control register:
cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
with this we disable vm86 extension, virtual interrupts, timestamp (RDTSC can only be executed with the highest privilege)
and debug extension. After this we reload the Glolbal Descripto Table and Interrupt Descriptor table with the:
switch_to_new_gdt(cpu);
loadsegment(fs, 0);
load_current_idt();
198
Linux Inside
After this we setup array of the Thread-Local Storage Descriptors, configure NX and load CPU microcode. Now is time to
setup and load per-cpu Task State Segements. We are going in a loop through the all exception stack which is
N_EXCEPTION_STACKS or 4 and fill it with Interrupt Stack Tables :
if (!oist->ist[0]) {
char *estacks = per_cpu(exception_stacks, cpu);
for (v = 0; v < N_EXCEPTION_STACKS; v++) {
estacks += exception_stack_sizes[v];
oist->ist[v] = t->x86_tss.ist[v] =
(unsigned long)estacks;
if (v == DEBUG_STACK-1)
per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
}
}
As we have filled Task State Segements with the Interrupt Stack Tables we can set TSS descriptor for the current
processor and load it with the:
set_tss_desc(cpu, t);
load_TR_desc();
where set_tss_desc macro from the arch/x86/include/asm/desc.h writes given descriptor to the Global Descriptor Table of
the given processor:
and load_TR_desc macro expands to the ltr or Load Task Register instruction:
In the end of the trap_init function we can see the following code:
199
Linux Inside
Here we copy idt_table to the nmi_dit_table and setup exception handlers for the #DB or Debug exception and #BR or
Breakpoint exception . You can remember that we already set these interrupt gates in the previous part, so why do we need
to setup it again? We setup it again because when we initialized it before in the early_trap_init function, the Task State
Segement was not ready yet, but now it is ready after the call of the cpu_init function.
Conclusion
It is the end of the fourth part about interrupts and interrupt handling in the Linux kernel. We saw the initialization of the
Task State Segment in this part and initialization of the different interrupt handlers as Divide Error , Page Fault excetpion
and etc. You can noted that we saw just initialization stuf, and will dive into details about handlers for these exceptions. In
the next part we will start to do it.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
page fault
Interrupt Descriptor Table
Tracing
cr2
RCU
thiscpu* operations
kmemcheck
prefetchw
3DNow
CPU caches
VFS
Linux kernel memory management
Fix-Mapped Addresses and ioremap
Extended Industry Standard Architecture
INT isntruction
INTO
BOUND
opcode
control register
x87 FPU
MCE exception
SIMD
cpumasks and bitmaps
NX
Task State Segment
Previous part
200
Linux Inside
The idtentry macro does following preparation before an actual exception handler ( do_divide_error for the divide_error ,
do_overflow for the overflow and etc.) will get control. In another words the idtentry macro allocates place for the
registers (pt_regs structure) on the stack, pushes dummy error code for the stack consistency if an interrupt/exception has
no error code, checks the segment selector in the cs segment register and switches depends on the previous
state(userspace or kernelspace). After all of these preparations it makes a call of an actual interrupt/exception handler:
After an exception handler will finish its work, the idtentry macro restores stack and general purpose registers of an
interrupted task and executes iret instruction:
ENTRY(paranoid_exit)
...
...
...
RESTORE_EXTRA_REGS
RESTORE_C_REGS
REMOVE_PT_GPREGS_FROM_STACK 8
INTERRUPT_RETURN
END(paranoid_exit)
201
Linux Inside
More about the idtentry macro you can read in the thirt part of the http://0xax.gitbooks.io/linuxinsides/content/interrupts/interrupts-3.html chapter. Ok, now we saw the preparation before an exception handler will be
executed and now time to look on the handlers. First of all let's look on the following handlers:
divide_error
overflow
invalid_op
coprocessor_segment_overrun
invalid_TSS
segment_not_present
stack_segment
alignment_check
All these handlers defined in the arch/x86/kernel/traps.c source code file with the DO_ERROR macro:
Note on the ## tokens. This is special feature - GCC macro Concatenation which concatenates two given strings. For
example, first DO_ERROR in our example will expands to the:
We can see that all functions which are generated by the DO_ERROR macro just make a call of the do_error_trap function
202
Linux Inside
Trap handlers
The do_error_trap function starts and ends from the two following functions:
from the include/linux/context_tracking.h. The context tracking in the Linux kernel subsystem which provide kernel
boundaries probes to keep track of the transitions between level contexts with two basic initial contexts: user or kernel .
The exception_enter function checks that context tracking is enabled. After this if it is enabled, the exception_enter reads
previous context and compares it with the CONTEXT_KERNEL . If the previous context is user , we call context_tracking_exit
function from the kernel/context_tracking.c which inform the context tracking subsystem that a processor is exiting user
mode and entering the kernel mode:
if (!context_tracking_is_enabled())
return 0;
prev_ctx = this_cpu_read(context_tracking.state);
if (prev_ctx != CONTEXT_KERNEL)
context_tracking_exit(prev_ctx);
return prev_ctx;
If previous context is non user , we just return it. The pre_ctx has enum ctx_state type which defined in the
include/linux/context_tracking_state.h and looks as:
enum ctx_state {
CONTEXT_KERNEL = 0,
CONTEXT_USER,
CONTEXT_GUEST,
} state;
The second function is exception_exit defined in the same include/linux/context_tracking.h file and checks that context
tracking is enabled and call the contert_tracking_enter function if the previous context was user :
The context_tracking_enter function informs the context tracking subsystem that a processor is going to enter to the user
mode from the kernel mode. We can see the following code between the exception_enter and exception_exit :
203
Linux Inside
First of all it calls the notify_die function which defined in the kernel/notifier.c. To get notified for kernel panic, kernel oops,
Non-Maskable Interrupt or other events the caller needs to insert itself in the notify_die chain and the notify_die function
does it. The Linux kernel has special mechanism that allows kernel to ask when something happens and this mechanism
called notifiers or notifier chains . This mechanism used for example for the USB hotplug events (look on the
drivers/usb/core/notify.c), for the memory hotplug (look on the include/linux/memory.h, the hotplug_memory_notifier macro
and etc...), system reboots and etc. A notifier chain is thus a simple, singly-linked list. When a Linux kernel subsystem
wants to be notified of specific events, it fills out a special notifier_block structure and passes it to the
notifier_chain_register function. An event can be sent with the call of the notifier_call_chain function. First of all the
notify_die function fills die_args structure with the trap number, trap string, registers and other values:
and returns the result of the atomic_notifier_call_chain function with the die_chain :
static ATOMIC_NOTIFIER_HEAD(die_chain);
return atomic_notifier_call_chain(&die_chain, val, &args);
which just expands to the atomit_notifier_head structure that contains lock and notifier_block :
struct atomic_notifier_head {
spinlock_t lock;
struct notifier_block __rcu *head;
};
The atomic_notifier_call_chain function calls each function in a notifier chain in turn and returns the value of the last
notifier function called. If the notify_die in the do_error_trap does not return NOTIFY_STOP we execute conditional_sti
function from the arch/x86/kernel/traps.c that checks the value of the interrupt flag and enables interrupt depends on it:
more about local_irq_enable macro you can read in the second part of this chapter. The next and last call in the
do_error_trap is the do_trap function. First of all the do_trap function defined the tsk variable which has trak_struct
type and represents the current interrupted process. After the definition of the tsk , we can see the call of the
do_trap_no_signal function:
204
Linux Inside
if (v8086_mode(regs)) {
...
}
if (!user_mode(regs)) {
...
}
return -1;
We will not consider first case because the long mode does not support the Virtual 8086 mode. In the second case we
invoke fixup_exception function which will try to recover a fault and die if we can't:
if (!fixup_exception(regs)) {
tsk->thread.error_code = error_code;
tsk->thread.trap_nr = trapnr;
die(str, regs, error_code);
}
The die function defined in the arch/x86/kernel/dumpstack.c source code file, prints useful information about stack,
registers, kernel modules and caused kernel oops. If we came from the userspace the do_trap_no_signal function will
return -1 and the execution of the do_trap function will continue. If we passed through the do_trap_no_signal function and
did not exit from the do_trap after this, it means that previous context was - user . Most exceptions caused by the
processor are interpreted by Linux as error conditions, for example division by zero, invalid opcode and etc. When an
exception occurs the Linux kernel sends a signal to the interrupted process that caused the exception to notify it of an
incorrect condition. So, in the do_trap function we need to send a signal with the given number ( SIGFPE for the divide error,
SIGILL for the overflow exception and etc...). First of all we save error code and vector number in the current interrupts
tsk->thread.error_code = error_code;
tsk->thread.trap_nr = trapnr;
After this we make a check do we need to print information about unhandled signals for the interrupted process. We check
that show_unhandled_signals variable is set, that unhandled_signal function from the kernel/signal.c will return unhandled
signal(s) and printk rate limit:
#ifdef CONFIG_X86_64
if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
printk_ratelimit()) {
pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
tsk->comm, tsk->pid, str,
regs->ip, regs->sp, error_code);
print_vma_addr(" in ", regs->ip);
pr_cont("\n");
}
#endif
205
Linux Inside
This is the end of the do_trap . We just saw generic implementation for eight different exceptions which are defined with the
DO_ERROR macro. Now let's look on another exception handlers.
Double fault
The next exception is #DF or Double fault . This exception occurrs when the processor detected a second exception while
calling an exception handler for a prior exception. We set the trap gate for this exception in the previous part:
Note that this exception runs on the DOUBLEFAULT_STACK Interrupt Stack Table which has index - 1 :
#define DOUBLEFAULT_STACK 1
The double_fault is handler for this exception and defined in the arch/x86/kernel/traps.c. The double_fault handler starts
from the definition of two variables: string that describes excetpion and interrupted process, as other exception handlers:
The handler of the double fault exception splitted on two parts. The first part is the check which checks that a fault is a nonIST fault on the espfix64 stack. Actually the iret instruction restores only the bottom 16 bits when returning to a 16 bit
segment. The espfix feature solves this problem. So if the non-IST fault on the espfix64 stack we modify the stack to
make it look like General Protection Fault :
In the second case we do almost the same that we did in the previous excetpion handlers. The first is the call of the
ist_enter function that discards previous context, user in our case:
ist_enter(regs);
And after this we fill the interrupted process with the vector number of the Double fault excetpion and error code as we did
it in the previous handlers:
tsk->thread.error_code = error_code;
tsk->thread.trap_nr = X86_TRAP_DF;
Next we print useful information about the double fault (PID number, registers content):
#ifdef CONFIG_DOUBLEFAULT
df_debug(regs, error_code);
#endif
206
Linux Inside
And die:
for (;;)
die(str, regs, error_code);
That's all.
BUG_ON(use_eager_fpu());
When we switch into a task or interrupt we may avoid loading the FPU state. If a task will use it, we catch Device not
Available exception exception. If we loading the FPU state during task switching, the FPU is eager. In the next step we
check cr0 control register on the EM flag which can show us is x87 floating point unit present (flag clear) or not (flag set):
#ifdef CONFIG_MATH_EMULATION
if (read_cr0() & X86_CR0_EM) {
struct math_emu_info info = { };
conditional_sti(regs);
info.regs = regs;
math_emulate(&info);
exception_exit(prev_state);
return;
}
#endif
If the x87 floating point unit not presented, we enable interrupts with the conditional_sti , fill the math_emu_info (defined in
the arch/x86/include/asm/math_emu.h) structure with the registers of an interrupt task and call math_emulate function from
the arch/x86/math-emu/fpu_entry.c. As you can understand from function's name, it emulates X87 FPU unit (more about the
Implementation of some exception handlers
207
Linux Inside
x87 we will know in the special chapter). In other way, if X86_CR0_EM flag is clear which means that x87 FPU unit is
presented, we call the fpu__restore function from the arch/x86/kernel/fpu/core.c which copies the FPU registers from the
fpustate to the live hardware registers. After this FPU instructions can be used:
fpu__restore(¤t->thread.fpu);
prev_state = exception_enter();
...
exception_exit(prev_state);
After this we enable interrupts if they were disabled and check that we came from the Virtual 8086 mode:
conditional_sti(regs);
if (v8086_mode(regs)) {
local_irq_enable();
handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
goto exit;
}
As long mode does not support this mode, we will not consider exception handling for this case. In the next step check that
previous mode was kernel mode and try to fix the trap. If we can't fix the current general protection fault exception we fill the
interrupted process with the vector number and error code of the exception and add it to the notify_die chain:
if (!user_mode(regs)) {
if (fixup_exception(regs))
goto exit;
tsk->thread.error_code = error_code;
tsk->thread.trap_nr = X86_TRAP_GP;
if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
die("general protection fault", regs, error_code);
goto exit;
}
If we can fix exception we go to the exit label which exits from exception state:
exit:
exception_exit(prev_state);
208
Linux Inside
If we came from user mode we send SIGSEGV signal to the interrupted process from user mode as we did it in the do_trap
function:
That's all.
Conclusion
It is the end of the fifth part of the Interrupts and Interrupt Handling chapter and we saw implementation of some interrupt
handlers in this part. In the next part we will continue to dive into interrupt and exception handlers and will see handler for
the Non-Maskable Interrupts, handling of the math coprocessor and SIMD coprocessor exceptions and many many more.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
Interrupt descriptor Table
iret instruction
GCC macro Concatenation
kernel panic
kernel oops
Non-Maskable Interrupt
hotplug
interrupt flag
long mode
signal
printk
coprocessor
SIMD
Interrupt Stack Table
PID
x87 FPU
control register
MMX
Previous part
209
Linux Inside
in the trap_init function which defined in the arch/x86/kernel/traps.c source code file. In the previous parts we saw that
entry points of the all interrupt handlers are defined with the:
macro from the arch/x86/entry/entry_64.S assembly source code file. But the handler of the Non-Maskable interrupts is not
defined with this macro. It has own entry point:
ENTRY(nmi)
...
...
...
END(nmi)
210
Linux Inside
in the same arch/x86/entry/entry_64.S assembly file. Lets dive into it and will try to understand how Non-Maskable interrupt
handler works. The nmi handlers starts from the call of the:
PARAVIRT_ADJUST_EXCEPTION_FRAME
macro but we will not dive into details about it in this part, because this macro related to the Paravirtualization stuff which
we will see in another chapter. After this save the content of the rdx register on the stack:
pushq %rdx
And allocated check that cs was not the kernel segment when an non-maskable interrupt occurs:
The __KERNEL_CS macro defined in the arch/x86/include/asm/segment.h and represented second descriptor in the Global
Descriptor Table:
#define GDT_ENTRY_KERNEL_CS 2
#define __KERNEL_CS (GDT_ENTRY_KERNEL_CS*8)
more about GDT you can read in the second part of the Linux kernel booting process chapter. If cs is not kernel segment,
it means that it is not nested NMI and we jump on the first_nmi label. Let's consider this case. First of all we put address
of the current stack pointer to the rdx and pushes 1 to the stack in the first_nmi label:
first_nmi:
movq (%rsp), %rdx
pushq $1
Why do we push 1 on the stack? As the comment says: We allow breakpoints in NMIs . On the x86_64, like other
architectures, the CPU will not execute another NMI until the first NMI is complete. A NMI interrupt finished with the iret
instruction like other interrupts and exceptions do it. If the NMI handler triggers either a page fault or breakpoint or another
exception which are use iret instruction too. If this happens while in NMI context, the CPU will leave NMI context and a
new NMI may come in. The iret used to return from those exceptions will re-enable NMIs and we will get nested nonmaskable interrupts. The problem the NMI handler will not return to the state that it was, when the exception triggered, but
instead it will return to a state that will allow new NMIs to preempt the running NMI handler. If another NMI comes in before
the first NMI handler is complete, the new NMI will write all over the preempted NMIs stack. We can have nested NMIs
where the next NMI is using the top of the stack of the previous NMI . It means that we cannot execute it because a nested
non-maskable interrupt will corrupt stack of a previous non-maskable interrupt. That's why we have allocated space on the
stack for temporary variable. We will check this variable that it was set when a previous NMI is executing and clear if it is
not nested NMI . We push 1 here to the previously allocated space on the stack to denote that a non-maskable interrupt
executed currently. Remember that when and NMI or another exception occurs we have the following stack frame:
+------------------------+
| SS |
| RSP |
| RFLAGS |
| CS |
| RIP |
+------------------------+
211
Linux Inside
and also an error code if an exception has it. So, after all of these manipulations our stack frame will look like this:
+------------------------+
| SS |
| RSP |
| RFLAGS |
| CS |
| RIP |
| RDX |
| 1 |
+------------------------+
and pushes the copy of the original stack frame after the allocated space:
.rept 5
pushq 11*8(%rsp)
.endr
with the .rept assembly directive. We need in the copy of the original stack frame. Generally we need in two copies of the
interrupt stack. First is copied interrupts stack: saved stack frame and copied stack frame. Now we pushes original stack
frame to the saved stack frame which locates after the just allocated 40 bytes ( copied stack frame). This stack frame is
used to fixup the copied stack frame that a nested NMI may change. The second - copied stack frame modified by any
nested NMIs to let the first NMI know that we triggered a second NMI and we should repeat the first NMI handler. Ok, we
have made first copy of the original stack frame, now time to make second copy:
After all of these manipulations our stack frame will be like this:
+-------------------------+
| original SS |
| original Return RSP |
| original RFLAGS |
| original CS |
| original RIP |
+-------------------------+
| temp storage for rdx |
+-------------------------+
| NMI executing variable |
+-------------------------+
| copied SS |
| copied Return RSP |
| copied RFLAGS |
| copied CS |
| copied RIP |
+-------------------------+
| Saved SS |
| Saved Return RSP |
| Saved RFLAGS |
| Saved CS |
| Saved RIP |
212
Linux Inside
+-------------------------+
After this we push dummy error code on the stack as we did it already in the previous exception handlers and allocate
space for the general purpose registers on the stack:
pushq $-1
ALLOC_PT_GPREGS_ON_STACK
We already saw implementation of the ALLOC_PT_GREGS_ON_STACK macro in the third part of the interrupts chapter. This macro
defined in the arch/x86/entry/calling.h and yet another allocates 120 bytes on stack for the general purpose registers, from
the rdi to the r15 :
After space allocation for the general registers we can see call of the paranoid_entry :
call paranoid_entry
We can remember from the previous parts this label. It pushes general purpose registers on the stack, reads MSR_GS_BASE
Model Specific register and checks its value. If the value of the MSR_GS_BASE is negative, we came from the kernel mode
and just return from the paranoid_entry , in other way it means that we came from the usermode and need to execute
swapgs instruction which will change user gs with the kernel gs :
ENTRY(paranoid_entry)
cld
SAVE_C_REGS 8
SAVE_EXTRA_REGS 8
movl $1, %ebx
movl $MSR_GS_BASE, %ecx
rdmsr
testl %edx, %edx
js 1f
SWAPGS
xorl %ebx, %ebx
1: ret
END(paranoid_entry)
Note that after the swapgs instruction we zeroed the ebx register. Next time we will check content of this register and if we
executed swapgs than ebx must contain 0 and 1 in other way. In the next step we store value of the cr2 control register
to the r12 register, because the NMI handler can cause page fault and corrupt the value of this control register:
Now time to call actual NMI handler. We push the address of the pt_regs to the rdi , error code to the rsi and call the
do_nmi handler:
213
Linux Inside
We will back to the do_nmi little later in this part, but now let's look what occurs after the do_nmi will finish its execution.
After the do_nmi handler will be finished we check the cr2 register, because we can got page fault during do_nmi
performed and if we got it we restore original cr2 , in other way we jump on the label 1 . After this we test content of the
ebx register (remember it must contain 0 if we have used swapgs instruction and 1 if we didn't use it) and execute
SWAPGS_UNSAFE_STACK if it contains 1 or jump to the nmi_restore label. The SWAPGS_UNSAFE_STACK macro just expands to the
swapgs instruction. In the nmi_restore label we restore general purpose registers, clear allocated space on the stack for
this registers clear our temporary variable and exit from the interrupt handler with the INTERRUPT_RETURN macro:
where INTERRUPT_RETURN is defined in the arch/x86/include/irqflags.h and just expands to the iret instruction. That's all.
Now let's consider case when another NMI interrupt occurred when previous NMI interrupt didn't finish its execution. You
can remember from the beginning of this part that we've made a check that we came from userspace and jump on the
first_nmi in this case:
Note that in this case it is first NMI every time, because if the first NMI catched page fault, breakpoint or another exception
it will be executed in the kernel mode. If we didn't come from userspace, first of all we test our temporary variable:
and if it is set to 1 we jump to the nested_nmi label. If it is not 1 , we test the IST stack. In the case of nested NMIs we
check that we are above the repeat_nmi . In this case we ignore it, in other way we check that we above than
end_repeat_nmi and jump on the nested_nmi_out label.
Now let's look on the do_nmi exception handler. This function defined in the arch/x86/kernel/nmi.c source code file and
takes two parameters:
address of the pt_regs ;
error code.
as all exception handlers. The do_nmi starts from the call of the nmi_nesting_preprocess function and ends with the call of
the nmi_nesting_postprocess . The nmi_nesting_preprocess function checks that we likely do not work with the debug stack
and if we on the debug stack set the update_debug_stack per-cpu variable to 1 and call the debug_stack_set_zero function
from the arch/x86/kernel/cpu/common.c. This function increases the debug_stack_use_ctr per-cpu variable and loads new
Interrupt Descriptor Table :
214
Linux Inside
The nmi_nesting_postprocess function checks the update_debug_stack per-cpu variable which we set in the
nmi_nesting_preprocess and resets debug stack or in another words it loads origin Interrupt Descriptor Table . After the
call of the nmi_nesting_preprocess function, we can see the call of the nmi_enter in the do_nmi . The nmi_enter increases
lockdep_recursion field of the interrupted process, update preempt counter and informs the RCU subsystem about NMI .
There is also nmi_exit function that does the same stuff as nmi_enter , but vice-versa. After the nmi_enter we increase
__nmi_count in the irq_stat structure and call the default_do_nmi function. First of all in the default_do_nmi we check the
address of the previous nmi and update address of the last nmi to the actual:
if (regs->ip == __this_cpu_read(last_nmi_rip))
b2b = true;
else
__this_cpu_write(swallow_nmi, false);
__this_cpu_write(last_nmi_rip, regs->ip);
reason = x86_platform.get_nmi_reason();
if (reason & NMI_REASON_MASK) {
if (reason & NMI_REASON_SERR)
pci_serr_error(reason, regs);
else if (reason & NMI_REASON_IOCHK)
io_check_error(reason, regs);
__this_cpu_add(nmi_stats.external, 1);
return;
}
That's all.
defined in the arch/x86/kernel/traps.c. The do_bounds handler starts with the call of the exception_enter function and ends
with the call of the exception_exit :
prev_state = exception_enter();
if (notify_die(DIE_TRAP, "bounds", regs, error_code,
X86_TRAP_BR, SIGSEGV) == NOTIFY_STOP)
215
Linux Inside
goto exit;
...
...
...
exception_exit(prev_state);
return;
After we have got the state of the previous context, we add the exception to the notify_die chain and if it will return
NOTIFY_STOP we return from the exception. More about notify chains and the context tracking functions you can read in
the previous part. In the next step we enable interrupts if they were disabled with the contidional_sti function that checks
IF flag and call the local_irq_enable depends on its value:
conditional_sti(regs);
if (!user_mode(regs))
die("bounds", regs, error_code);
and check that if we didn't came from user mode we send SIGSEGV signal with the die function. After this we check is MPX
enabled or not, and if this feature is disabled we jump on the exit_trap label:
if (!cpu_feature_enabled(X86_FEATURE_MPX)) {
goto exit_trap;
}
where we execute `do_trap` function (more about it you can find in the previous part):
```C
exit_trap:
do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, NULL);
exception_exit(prev_state);
If MPX feature is enabled we check the BNDSTATUS with the get_xsave_field_ptr function and if it is zero, it means that the
MPX was not responsible for this exception:
bndcsr = get_xsave_field_ptr(XSTATE_BNDCSR);
if (!bndcsr)
goto exit_trap;
After all of this, there is still only one way when MPX is responsible for this exception. We will not dive into the details about
Intel Memory Protection Extensions in this part, but will see it in another chapter.
of the math_error function from the same source code file but pass different vector number. The do_coprocessor_error
passes X86_TRAP_MF vector number to the math_error :
216
Linux Inside
dotraplinkage void
do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
{
enum ctx_state prev_state;
prev_state = exception_enter();
math_error(regs, error_code, X86_TRAP_XF);
exception_exit(prev_state);
}
First of all the math_error function defines current interrupted task, address of its fpu, string which describes an exception,
add it to the notify_die chain and return from the exception handler if it will return NOTIFY_STOP :
After this we check that we are from the kernel mode and if yes we will try to fix an excetpion with the fixup_exception
function. If we cannot we fill the task with the exception's error code and vector number and die:
if (!user_mode(regs)) {
if (!fixup_exception(regs)) {
task->thread.error_code = error_code;
task->thread.trap_nr = trapnr;
die(str, regs, error_code);
}
return;
}
If we came from the user mode, we save the fpu state, fill the task structure with the vector number of an exception and
siginfo_t with the number of signal, errno , the address where exception occurred and signal code:
fpu__save(fpu);
task->thread.trap_nr = trapnr;
task->thread.error_code = error_code;
info.si_signo = SIGFPE;
info.si_errno = 0;
info.si_addr = (void __user *)uprobe_get_trap_addr(regs);
info.si_code = fpu__exception_code(fpu, trapnr);
if (!info.si_code)
return;
217
Linux Inside
That's all.
Conclusion
It is the end of the sixth part of the Interrupts and Interrupt Handling chapter and we saw implementation of some exception
handlers in this part, like non-maskable interrupt, SIMD and x87 FPU floating point exception. Finally we have finsihed with
the trap_init function in this part and will go ahead in the next part. The next our point is the external interrupts and the
early_irq_init function from the init/main.c.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
General Protection Fault
opcode
Non-Maskable
BOUND instruction
CPU socket
Interrupt Descriptor Table
Interrupt Stack Table
Paravirtualization
.rept
SIMD
Coprocessor
x86_64
iret
page fault
breakpoint
Global Descriptor Table
stack frame
Model Specific regiser
percpu
RCU
MPX
x87 FPU
Previous part
218
Linux Inside
table that is located at fixed address in the memory. After the interrupt is handled processor resumes the interrupted
process. At the boot/initialization time, the Linux kernel identifies all devices in the machine, and appropriate interrupt
handlers are loaded into the interrupt table. As we saw in the previous parts, most exceptions are handled simply by the
sending a Unix signal to the interrupted process. That's why kernel is can handle an exception quickly. Unfortunatelly we
can not use this approach for the external handware interrupts, because often they arrive after (and sometimes long after)
the process to which they are related has been suspended. So it would make no sense to send a Unix signal to the current
process. External interrupt handling depends on the type of an interrupt:
I/O interrupts;
Timer interrupts;
Interprocessor interrupts.
I will try to describe all types of interrupts in this book.
Generally, a handler of an I/O interrupt must be flexible enough to service several devices at the same time. For exmaple
in the PCI bus architecture several devices may share the same IRQ line. In the simplest way the Linux kernel must do
following thing when an I/O interrupt occured:
Save the value of an IRQ and the register's contents on the kernel stack;
Send an acknowledgment to the hardware controller which is servicing the IRQ line;
Execute the interrupt service routine (next we will call it ISR ) which is associated with the device;
Restore registers and return from an interrupt;
Ok, we know a little theory and now let's start with the early_irq_init function. The implementation of the early_irq_init
function is in the kernel/irq/irqdesc.c. This function make early initialziation of the irq_desc structure. The irq_desc
structure is the foundation of interrupt management code in the Linux kernel. An array of this structure, which has the same
name - irq_desc , keeps track of every interrupt request source in the Linux kernel. This structure defined in the
include/linux/irqdesc.h and as you can note it depends on the CONFIG_SPARSE_IRQ kernel configuration option. This kernel
configuration option enables support for sparse irqs. The irq_desc structure contains many different fiels:
irq_common_data - per irq and chip data passed down to chip functions;
status_use_accessors - contains status of the interrupt source which is can be combination of of the values from the
enum from the include/linux/irq.h and different macros which are defined in the same source code file;
kstat_irqs - irq stats per-cpu;
handle_irq - highlevel irq-events handler;
action - identifies the interrupt service routines to be invoked when the IRQ occurs;
219
Linux Inside
declaration of the following variables: irq descriptors counter, loop counter, memory node and the irq_desc descriptor:
The node is an online NUMA node which depends on the MAX_NUMNODES value which depends on the CONFIG_NODES_SHIFT
kernel configuration parameter:
As I already wrote, implementation of the first_online_node macro depends on the MAX_NUMNODES value:
The node_states is the enum which defined in the include/linux/nodemask.h and represent the set of the states of a node.
In our case we are searching an online node and it will be 0 if MAX_NUMNODES is one or zero. If the MAX_NUMNODES is greater
than one, the node_states[N_ONLINE] will return 1 and the first_node macro will be expands to the call of the
__first_node function which will return minimal or the first online node:
220
Linux Inside
More about this will be in the another chapter about the NUMA . The next step after the declaration of these local variables is
the call of the:
init_irq_default_affinity();
function. The init_irq_default_affinity function defined in the same source code file and depends on the CONFIG_SMP
kernel configuration option allocates a given cpumask structure (in our case it is the irq_default_affinity ):
#if defined(CONFIG_SMP)
cpumask_var_t irq_default_affinity;
static void __init init_irq_default_affinity(void)
{
alloc_cpumask_var(&irq_default_affinity, GFP_NOWAIT);
cpumask_setall(irq_default_affinity);
}
#else
static void __init init_irq_default_affinity(void)
{
}
#endif
We know that when a hardware, such as disk controller or keyboard, needs attention from the processor, it throws an
interrupt. The interrupt tells to the processor that something has happened and that the processor should interrupt current
process and handle an incoming event. In order to prevent mutliple devices from sending the same interrupts, the IRQ
system was established where each device in a computer system is assigned its own special IRQ so that its interrupts are
unique. Linux kernel can assign certain IRQs to specific processors. This is known as SMP IRQ affinity , and it allows you
control how your system will respond to various hardware events (that's why it has certain implementation only if the
CONFIG_SMP kernel configuration option is set). After we allocated irq_default_affinity cpumask, we can see printk
output:
The NR_IRQS is the maximum number of the irq descriptors or in another words maximum number of interrupts. Its value
depends on the state of the COFNIG_X86_IO_APIC kernel configuration option. If the CONFIG_X86_IO_APIC is not set and the
Linux kernel uses an old PIC chip, the NR_IRQS is:
#define NR_IRQS_LEGACY 16
#ifdef CONFIG_X86_IO_APIC
...
...
221
Linux Inside
...
#else
# define NR_IRQS NR_IRQS_LEGACY
#endif
In other way, when the CONFIG_X86_IO_APIC kernel configuration option is set, the NR_IRQS depends on the amount of the
processors and amount of the interrupt vectors:
We remember from the previous parts, that the amount of processors we can set during Linux kernel configuration process
with the CONFIG_NR_CPUS configuration option:
In the first case ( CPU_VECTOR_LIMIT > IO_APIC_VECTOR_LIMIT ), the NR_IRQS will be 4352 , in the second case ( CPU_VECTOR_LIMIT
< IO_APIC_VECTOR_LIMIT ), the NR_IRQS will be 768 . In my case the NR_CPUS is 8 as you can see in the my configuration, the
CPU_VECTOR_LIMIT is 512 and the IO_APIC_VECTOR_LIMIT is 4096 . So NR_IRQS for my configuration is 4352 :
In the next step we assign array of the IRQ descriptors to the irq_desc variable which we defined in the start of the
early_irq_init function and cacluate count of the irq_desc array with the ARRAY_SIZE macro:
222
Linux Inside
desc = irq_desc;
count = ARRAY_SIZE(irq_desc);
The irq_desc array defined in the same source code file and looks like:
The irq_desc is array of the irq descriptors. It has three already initialized fields:
handle_irq - as I already wrote above, this field is the highlevel irq-event handler. In our case it initialized with the
handle_bad_irq function that defined in the kernel/irq/handle.c source code file and handles spurious and unhandled
irqs;
depth - 0 if the IRQ line is enabled and a positive value if it has been disabled at least once;
lock - A spin lock used to serialize the accesses to the IRQ descriptor.
As we calculated count of the interrupts and initialized our irq_desc array, we start to fill descriptors in the loop:
We are going through the all interrupt descriptors and do the following things:
First of all we allocate percpu variable for the irq kernel statistic with the alloc_percpu macro. This macro allocates one
instance of an object of the given type for every processor on the system. You can access kernel statistic from the
userspace via /proc/stat :
~$ cat /proc/stat
cpu 207907 68 53904 5427850 14394 0 394 0 0 0
cpu0 25881 11 6684 679131 1351 0 18 0 0 0
cpu1 24791 16 5894 679994 2285 0 24 0 0 0
cpu2 26321 4 7154 678924 664 0 71 0 0 0
cpu3 26648 8 6931 678891 414 0 244 0 0 0
...
...
...
Where the sixth column is the servicing interrupts. After this we allocate cpumask for the given irq descriptor affinity and
initialize the spinlock for the given interrupt descriptor. After this before the critical section, the lock will be aqcuired with a
call of the raw_spin_lock and unlocked with the call of the raw_spin_unlock . In the next step we call the lockdep_set_class
macro which set the Lock validator irq_desc_lock_class class for the lock of the given interrupt descriptor. More about
lockdep , spinlock and other synchronization primitives will be described in the separate chapter.
In the end of the loop we call the desc_set_defaults function from the kernel/irq/irqdesc.c. This function takes four
parameters:
number of a irq;
Dive into external hardware interrupts
223
Linux Inside
interrupt descriptor;
online NUMA node;
owner of interrupt descriptor. Interrupt descriptors can be allocated from modules. This field is need to proved refcount
on the module which provides the interrupts;
and fills the rest of the irq_desc fields. The desc_set_defaults function fills interrupt number, irq chip, platform-specific
per-chip private data for the chip methods, per-IRQ data for the irq_chip methods and MSI descriptor for the per irq and
irq chip data:
desc->irq_data.irq = irq;
desc->irq_data.chip = &no_irq_chip;
desc->irq_data.chip_data = NULL;
desc->irq_data.handler_data = NULL;
desc->irq_data.msi_desc = NULL;
...
...
...
The irq_data.chip structure provides general API like the irq_set_chip , irq_set_irq_type and etc, for the irq controller
drivers. You can find it in the kernel/irq/chip.c source code file.
After this we set the status of the accessor for the given descriptor and set disabled state of the interrupts:
...
...
...
irq_settings_clr_and_set(desc, ~0, _IRQ_DEFAULT_INIT_FLAGS);
irqd_set(&desc->irq_data, IRQD_IRQ_DISABLED);
...
...
...
In the next step we set the high level interrupt handlers to the handle_bad_irq which handles spurious and unhandled irqs
(as the hardware stuff is not initialized yet, we set this handler), set irq_desc.desc to 1 which means that an IRQ is
disabled, reset count of the unhandled interrupts and interrupts in general:
...
...
...
desc->handle_irq = handle_bad_irq;
desc->depth = 1;
desc->irq_count = 0;
desc->irqs_unhandled = 0;
desc->name = NULL;
desc->owner = owner;
...
...
...
After this we go through the all possible processor with the for_each_possible_cpu helper and set the kstat_irqs to zero
for the given interrupt descriptor:
for_each_possible_cpu(cpu)
*per_cpu_ptr(desc->kstat_irqs, cpu) = 0;
and call the desc_smp_init function from the kernel/irq/irqdesc.c that initializes NUMA node of the given interrupt descriptor,
sets default SMP affinity and clears the pending_mask of the given interrupt descriptor depends on the value of the
224
Linux Inside
In the end of the early_irq_init function we return the return value of the arch_early_irq_init function:
return arch_early_irq_init();
This function defined in the kernel/apic/vector.c and contains only one call of the arch_early_ioapic_init function from the
kernel/apic/io_apic.c. As we can understand from the arch_early_ioapic_init function's name, this function makes early
initialization of the I/O APIC. First of all it make a check of the number of the legacy interrupts wit the call of the
nr_legacy_irqs function. If we have no lagacy interrupts with the Intel 8259 programmable interrupt controller we set
io_apic_irqs to the 0xffffffffffffffff :
if (!nr_legacy_irqs())
io_apic_irqs = ~0UL;
After this we are going through the all I/O APICs and allocate space for the registers with the call of the
alloc_ioapic_saved_registers :
for_each_ioapic(i)
alloc_ioapic_saved_registers(i);
And in the end of the arch_early_ioapic_init function we are going through the all legacy irqs (from IRQ0 to IRQ15 ) in the
loop and allocate space for the irq_cfg which represents configuration of an irq on the given NUMA node:
That's all.
Sparse IRQs
We already saw in the beginning of this part that implementation of the early_irq_init function depends on the
CONFIG_SPARSE_IRQ kernel configuration option. Previously we saw implementation of the early_irq_init function when the
CONFIG_SPARSE_IRQ configuration option is not set, not let's look on the its implementation when this option is set.
Implementation of this function very similar, but little differ. We can see the same definition of variables and call of the
init_irq_default_affinity in the beginning of the early_irq_init function:
#ifdef CONFIG_SPARSE_IRQ
int __init early_irq_init(void)
225
Linux Inside
{
int i, initcnt, node = first_online_node;
struct irq_desc *desc;
init_irq_default_affinity();
...
...
...
}
#else
...
...
...
initcnt = arch_probe_nr_irqs();
The arch_probe_nr_irqs function defined in the arch/x86/kernel/apic/vector.c and calculates count of the pre-allocated irqs
and update nr_irqs with its number. But stop. Why there are pre-allocated irqs? There is alternative form of interrupts
called - Message Signaled Interrupts available in the PCI. Instead of assigning a fixed number of the interrupt request, the
device is allowed to record a message at a particular address of RAM, in fact, the display on the Local APIC. MSI permits a
device to allocate 1 , 2 , 4 , 8 , 16 or 32 interrupts and MSI-X permits a device to allocate up to 2048 interrupts. Now we
know that irqs can be pre-allocated. More about MSI will be in a next part, but now let's look on the arch_probe_nr_irqs
function. We can see the check which assign amount of the interrupt vectors for the each processor in the system to the
nr_irqs if it is greater and calculate the nr which represents number of MSI interrupts:
Take a look on the gsi_top variable. Each APIC is identified with its own ID and with the offset where its IRQ starts. It is
called GSI base or Global System Interrupt base. So the gsi_top represnters it. We get the Global System Interrupt
base from the MultiProcessor Configuration Table table (you can remember that we have parsed this table in the sixth part
of the Linux Kernel initialization process chapter).
After this we update the nr depends on the value of the gsi_top :
Update the nr_irqs if it less than nr and return the number of the legacy irqs:
The next after the arch_probe_nr_irqs is printing information about number of IRQs :
Dive into external hardware interrupts
226
Linux Inside
After this we do some checks that nr_irqs and initcnt values is not greater than maximum allowable number of irqs :
where IRQ_BITMAP_BITS is equal to the NR_IRQS if the CONFIG_SPARSE_IRQ is not set and NR_IRQS + 8196 in other way. In the
next step we are going over all interrupt descript which need to be allocated in the loop and allocate space for the
descriptor and insert to the irq_desc_tree radix tree:
In the end of the early_irq_init function we return the value of the call of the arch_early_irq_init function as we did it
already in the previous variant when the CONFIG_SPARSE_IRQ option was not set:
return arch_early_irq_init();
That's all.
Conclusion
It is the end of the seventh part of the Interrupts and Interrupt Handling chapter and we started to dive into external
hardware interrupts in this part. We saw early initialization of the irq_desc structure which represents description of an
external interrupt and contains information about it like list of irq actions, information about interrupt handler, interrupts's
owner, count of the unhandled interrupt and etc. In the next part we will continue to research external interrupts.
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
IRQ
numa
Enum type
227
Linux Inside
cpumask
percpu
spinlock
critical section
Lock validator
MSI
I/O APIC
Local APIC
Intel 8259
PIC
MultiProcessor Configuration Table
radix tree
dmesg
228
Linux Inside
kernel and represents an interrupt descriptor. In this part we will continue to dive into the initialization stuff which is related
to the external hardware interrupts.
Right after the call of the early_irq_init function in the init/main.c we can see the call of the init_IRQ function. This
function is architecture-specfic and defined in the arch/x86/kernel/irqinit.c. The init_IRQ function makes initialization of the
vector_irq percpu variable that defined in the same arch/x86/kernel/irqinit.c source code file:
...
DEFINE_PER_CPU(vector_irq_t, vector_irq) = {
[0 ... NR_VECTORS - 1] = -1,
};
...
and represents percpu array of the interrupt vector numbers. The vector_irq_t defined in the
arch/x86/include/asm/hw_irq.h and expands to the:
where NR_VECTORS is count of the vector number and as you can remember from the first part of this chapter it is 256 for
the x86_64:
So, in the start of the init_IRQ function we fill the vecto_irq percpu array with the vector number of the legacy interrupts:
This vector_irq will be used during the first steps of an external hardware interrupt handling in the do_IRQ function from
the arch/x86/kernel/irq.c:
229
Linux Inside
...
irq = __this_cpu_read(vector_irq[vector]);
if (!handle_irq(irq, regs)) {
...
...
...
}
exiting_irq();
...
...
return 1;
}
Why is legacy here? Actuall all interrupts handled by the modern IO-APIC controller. But these interrupts (from 0x30 to
0x3f ) by legacy interrupt-controllers like Programmable Interrupt Controller. If these interrupts are handled by the I/O
APIC then this vector space will be freed and re-used. Let's look on this code closer. First of all the nr_legacy_irqs defined
in the arch/x86/include/asm/i8259.h and just returns the nr_legacy_irqs field from the legacy_pic strucutre:
This structure defined in the same header file and represents non-modern programmable interrupts controller:
struct legacy_pic {
int nr_legacy_irqs;
struct irq_chip *chip;
void (*mask)(unsigned int irq);
void (*unmask)(unsigned int irq);
void (*mask_all)(void);
void (*restore_mask)(void);
void (*init)(int auto_eoi);
int (*irq_pending)(unsigned int irq);
void (*make_irq)(unsigned int irq);
};
Actuall default maximum number of the legacy interrupts represtented by the NR_IRQ_LEGACY macro from the
arch/x86/include/asm/irq_vectors.h:
#define NR_IRQS_LEGACY 16
In the loop we are accessing the vecto_irq per-cpu array with the per_cpu macro by the IRQ0_VECTOR + i index and write
the legacy vector number there. The IRQ0_VECTOR macro defined in the arch/x86/include/asm/irq_vectors.h header file and
expands to the 0x30 :
Why is 0x30 here? You can remember from the first part of this chapter that first 32 vector numbers from 0 to 31 are
reserved by the processor and used for the processing of architecture-defined exceptions and interrupts. Vector numbers
from 0x30 to 0x3f are reserved for the ISA. So, it means that we fill the vector_irq from the IRQ0_VECTOR which is equal
to the 32 to the IRQ0_VECTOR + 16 (before the 0x30 ).
230
Linux Inside
In the end of the init_IRQ functio we can see the call of the following function:
x86_init.irqs.intr_init();
from the arch/x86/kernel/x86_init.c source code file. If you have read chapter about the Linux kernel initialization process,
you can remember the x86_init structure. This structure contains a couple of files which are points to the function related
to the platform setup ( x86_64 in our case), for example resources - related with the memory resources, mpparse - related
with the parsing of the MultiProcessor Configuration Table table and etc.). As we can see the x86_init also contains the
irqs field which contains three following fields:
Now, we are interesting in the native_init_IRQ . As we can note, the name of the native_init_IRQ function contains the
native_ prefix which means that this function is architecture-specific. It defined in the arch/x86/kernel/irqinit.c and executes
general initialization of the Local APIC and initialization of the ISA irqs. Let's look on the implementation of the
native_init_IRQ function and will try to understand what occurs there. The native_init_IRQ function starts from the
x86_init.irqs.pre_vector_init();
As we can see above, the pre_vector_init points to the init_ISA_irqs function that defined in the same source code file
and as we can understand from the function's name, it makes initialization of the ISA related interrupts. The init_ISA_irqs
function starts from the definition of the chip variable which has a irq_chip type:
The irq_chip structure defined in the include/linux/irq.h header file and represents hardware interrupt chip descriptor. It
contains:
name - name of a device. Used in the /proc/interrupts :
$ cat /proc/interrupts
CPU0 CPU1 CPU2 CPU3 CPU4 CPU5 CPU6 CPU7
0: 16 0 0 0 0 0 0 0 IO-APIC 2-edge timer
1: 2 0 0 0 0 0 0 0 IO-APIC 1-edge i8042
8: 1 0 0 0 0 0 0 0 IO-APIC 8-edge rtc0
231
Linux Inside
and etc.
fields. Note that the irq_data structure represents set of the per irq chip data passed down to chip functions. It contains
mask - precomputed bitmask for accessing the chip registers, irq - interrupt number, hwirq - hardware interrupt number,
local to the interrupt domain chip low level interrupt hardware access and etc.
After this depends on the CONFIG_X86_64 and CONFIG_X86_LOCAL_APIC kernel configuration option call the init_bsp_APIC
function from the arch/x86/kernel/apic/apic.c:
This function makes initialization of the APIC of bootstrap processor (or processor which starts first). It starts from the
check that we found SMP config (read more about it in the sixth part of the Linux kernel initialization process chapter) and
the processor has APIC :
if (smp_found_config || !cpu_has_apic)
return;
In other way we return from this function. In the next step we call the clear_local_APIC function from the same source code
file that shutdowns the local APIC (more about it will be in the chapter about the Advanced Programmable Interrupt
Controller ) and enable APIC of the first processor by the setting unsigned int value to the APIC_SPIV_APIC_ENABLED :
value = apic_read(APIC_SPIV);
value &= ~APIC_VECTOR_MASK;
value |= APIC_SPIV_APIC_ENABLED;
apic_write(APIC_SPIV, value);
After we have enabled APIC for the bootstrap processor, we return to the init_ISA_irqs function and in the next step we
initalize legacy Programmable Interrupt Controller and set the legacy chip and handler for the each legacy irq:
legacy_pic->init(0);
for (i = 0; i < nr_legacy_irqs(); i++)
irq_set_chip_and_handler(i, chip, handle_level_irq);
Where can we find init function? The legacy_pic defined in the arch/x86/kernel/i8259.c and it is:
232
Linux Inside
The init_8259A function defined in the same source code file and executes initialization of the Intel 8259 `Programmable
Interrupt Controller (more about it will be in the separate chapter abot Programmable Interrupt Controllers and APIC ).
Now we can return to the native_init_IRQ function, after the init_ISA_irqs function finished its work. The next step is the
call of the apic_intr_init function that allocates special interrupt gates which are used by the SMP architecture for the
Inter-processor interrupt. The alloc_intr_gate macro from the arch/x86/include/asm/desc.h used for the interrupt
descriptor allocation allocation:
As we can see, first of all it expands to the call of the alloc_system_vector function that checks the given vector number in
the user_vectors bitmap (read previous part about it) and if it is not set in the user_vectors bitmap we set it. After this we
test that the first_system_vector is greater than given interrupt vector number and if it is greater we assign it:
if (!test_bit(vector, used_vectors)) {
set_bit(vector, used_vectors);
if (first_system_vector > vector)
first_system_vector = vector;
} else {
BUG();
}
We already saw the set_bit macro, now let's look on the test_bit and the first_system_vector . The first test_bit
macro defined in the arch/x86/include/asm/bitops.h and looks like this:
We can see the ternary operator here make a test with the gcc built-in function __builtin_constant_p tests that given vector
number ( nr ) is known at compile time. If you're feeling misunderstanding of the __builtin_constant_p , we can make
simple test:
#include <stdio.h>
#define PREDEFINED_VAL 1
int main() {
int i = 5;
printf("__builtin_constant_p(i) is %d\n", __builtin_constant_p(i));
233
Linux Inside
Now I think it must be clear for you. Let's get back to the test_bit macro. If the __builtin_constant_p will return non-zero,
we call constant_test_bit function:
What's the difference between two these functions and why do we need in two different functions for the same purpose? As
you already can guess main purpose is optimization. If we will write simple example with these functions:
#define CONST 25
int main() {
int nr = 24;
variable_test_bit(nr, (int*)0x10000000);
constant_test_bit(CONST, (int*)0x10000000)
return 0;
}
and will look on the assembly output of our example we will see followig assembly code:
pushq %rbp
movq %rsp, %rbp
movl $268435456, %esi
movl $25, %edi
call constant_test_bit
234
Linux Inside
pushq %rbp
movq %rsp, %rbp
subq $16, %rsp
movl $24, -4(%rbp)
movl -4(%rbp), %eax
movl $268435456, %esi
movl %eax, %edi
call variable_test_bit
for the variable_test_bit . These two code listings starts with the same part, first of all we save base of the current stack
frame in the %rbp register. But after this code for both examples is different. In the first example we put $268435456 (here
the $268435456 is our second parameter - 0x10000000 ) to the esi and $25 (our first parameter) to the edi register and
call constant_test_bit . We put functuin parameters to the esi and edi registers because as we are learning Linux kernel
for the x86_64 architecture we use System V AMD64 ABI calling convention. All is pretty simple. When we are using
predifined constant, the compiler can just substitute its value. Now let's look on the second part. As you can see here, the
compiler can not substitute value from the nr variable. In this case compiler must calcuate its offset on the programm's
stack frame. We substract 16 from the rsp register to allocate stack for the local variables data and put the $24 (value of
the nr variable) to the rbp with offset -4 . Our stack frame will be like this:
After this we put this value to the eax , so eax register now contains value of the nr . In the end we do the same that in the
first example, we put the $268435456 (the first parameter of the variable_test_bit function) and the value of the eax
(value of nr ) to the edi register (the second parameter of the variable_test_bit function ).
The next step after the apic_intr_init function will finish its work is the setting interrup gates from the
FIRST_EXTERNAL_VECTOR or 0x20 to the 0x256 :
i = FIRST_EXTERNAL_VECTOR;
#ifndef CONFIG_X86_LOCAL_APIC
#define first_system_vector NR_VECTORS
#endif
for_each_clear_bit_from(i, used_vectors, first_system_vector) {
set_intr_gate(i, irq_entries_start + 8 * (i - FIRST_EXTERNAL_VECTOR));
}
But as we are using the for_each_clear_bit_from helper, we set only non-initialized interrupt gates. After this we use the
same for_each_clear_bit_from helper to fill the non-filled interrupt gates in the interrupt table with the spurious_interrupt :
#ifdef CONFIG_X86_LOCAL_APIC
for_each_clear_bit_from(i, used_vectors, NR_VECTORS)
set_intr_gate(i, spurious_interrupt);
#endif
Where the spurious_interrupt function represent interrupt handler fro the spurious interrupt. Here the used_vectors is the
Initialization of external hardware interrupts structures
235
Linux Inside
unsigned long that contains already initialized interrupt gates. We already filled first 32 interrupt vectors in the trap_init
You can remember how we did it in the sixth part of this chapter.
In the end of the native_init_IRQ function we can see the following check:
First of all let's deal with the condition. The acpi_ioapic variable represents existence of I/O APIC. It defined in the
arch/x86/kernel/acpi/boot.c. This variable set in the acpi_set_irq_model_ioapic function that called during the processing
Multiple APIC Description Table . This occurs during initialization of the architecture-specific stuff in the
arch/x86/kernel/setup.c (more about it we will know in the other chapter about APIC). Note that the value of the
acpi_ioapic variable depends on the CONFIG_ACPI and CONFIG_X86_LOCAL_APIC Linux kernel configuration options. If these
#define acpi_ioapic 0
The second condition - !of_ioapic && nr_legacy_irqs() checks that we do not use Open Firmware I/O APIC and legacy
interrupt controller. We already know about the nr_legacy_irqs . The second is of_ioapic variable defined in the
arch/x86/kernel/devicetree.c and initialized in the dtb_ioapic_setup function that build information about APICs in the
devicetree. Note that of_ioapic variable depends on the CONFIG_OF Linux kernel configuration opiotn. If this option is not
set, the value of the of_ioapic will be zero too:
#ifdef CONFIG_OF
extern int of_ioapic;
...
...
...
#else
#define of_ioapic 0
...
...
...
#endif
setup_irq(2, &irq2);
function. First of all about the irq2 . The irq2 is the irqaction structure that defined in the arch/x86/kernel/irqinit.c source
code file and represents IRQ 2 line that is used to query devices connected cascade:
236
Linux Inside
Some time ago interrupt controller consisted of two chips and one was connected to second. The second chip that was
connected to the first chip via this IRQ 2 line. This chip serviced lines from 8 to 15 and after after this lines of the first
chip. So, for example Intel 8259A has following lines:
IRQ 0 - system time;
IRQ 1 - keyboard;
IRQ 2 - used for devices which are cascade connected;
IRQ 8 - RTC;
IRQ 9 - reserved;
IRQ 10 - reserved;
IRQ 11 - reserved;
IRQ 12 - ps/2 mouse;
IRQ 13 - coprocessor;
IRQ 14 - hard drive controller;
IRQ 1 - reserved;
IRQ 3 - COM2 and COM4 ;
IRQ 4 - COM1 and COM3 ;
IRQ 5 - LPT2 ;
IRQ 6 - drive controller;
IRQ 7 - LPT1 .
The setup_irq function defined in the kernel/irq/manage.c and takes two parameters:
vector number of an interrupt;
irqaction structure related with an interrupt.
This function initializes interrupt descriptor from the given vector number at the beginning:
chip_bus_lock(desc);
retval = __setup_irq(irq, desc, act);
chip_bus_sync_unlock(desc);
return retval;
Note that the interrupt descriptor is locked during __setup_irq function will work. The __setup_irq function makes many
different things: It creates a handler thread when a thread function is supplied and the interrupt does not nest into another
interrupt thread, sets the flags of the chip, fills the irqaction structure and many many more.
All of the above it creates /prov/vector_number directory and fills it, but if you are using modern computer all values will be
zero there:
$ cat /proc/irq/2/node
0
$cat /proc/irq/2/affinity_hint
00
cat /proc/irq/2/spurious
count 0
unhandled 0
last_unhandled 0 ms
237
Linux Inside
Conclusion
It is the end of the eighth part of the Interrupts and Interrupt Handling chapter and we continued to dive into external
hardware interrupts in this part. In the previous part we started to do it and saw early initialization of the IRQs . In this part
we already saw non-early interrupts initialization in the init_IRQ function. We saw initialization of the vector_irq per-cpu
array which is store vector numbers of the interrupts and will be used during interrupt handling and initialization of other
stuff which is related to the external hardware interrupts.
In the next part we will continue to learn interrupts handling related stuff and will see initialization of the softirqs .
If you will have any questions or suggestions write me a comment or ping me at twitter.
Please note that English is not my first language, And I am really sorry for any inconvenience. If you will find any
mistakes please send me PR to linux-internals.
Links
IRQ
percpu
x86_64
Intel 8259
Programmable Interrupt Controller
ISA
MultiProcessor Configuration Table
Local APIC
I/O APIC
SMP
Inter-processor interrupt
ternary operator
gcc
calling convention
PDF. System V Application Binary Interface AMD64
Call stack
Open Firmware
devicetree
RTC
Previous part
238
Linux Inside
Memory management
239
Linux Inside
time. No compilcated memory management is working yet. When the start_kernel function is called we will see the
transition to more complex data structures and techniques for memory management. For a good understanding of the
initialization process in the linux kernel we need to have a clear understanding of these techniques. This chapter will
provide an overview of the different parts of the linux kernel memory management framework and its API, starting from the
memblock .
Memblock
Memblock is one of the methods of managing memory regions during the early bootstrap period while the usual kernel
memory allocators are not up and running yet. Previously it was called Logical Memory Block , but with the patch by Yinghai
Lu, it was renamed to the memblock . As Linux kernel for x86_64 architecture uses this method. We already met memblock in
the Last preparations before the kernel entry point part. And now time to get acquainted with it closer. We will see how it is
implemented.
We will start to learn memblock from the data structures. Definitions of the all data structures can be found in the
include/linux/memblock.h header file.
The first structure has the same name as this part and it is:
struct memblock {
bool bottom_up;
phys_addr_t current_limit;
struct memblock_type memory; --> array of memblock_region
struct memblock_type reserved; --> array of memblock_region
#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
struct memblock_type physmem;
#endif
};
This structure contains five fields. First is bottom_up which allows allocating memory in bottom-up mode when it is true .
Next field is current_limit . This field describes the limit size of the memory block. The next three fields describe the type
of the memory block. It can be: reserved, memory and physical memory if the CONFIG_HAVE_MEMBLOCK_PHYS_MAP configuration
option is enabled. Now we see yet another data structure - memblock_type . Let's look at its definition:
struct memblock_type {
unsigned long cnt;
unsigned long max;
phys_addr_t total_size;
struct memblock_region *regions;
};
This structure provides information about memory type. It contains fields which describe the number of memory regions
which are inside the current memory block, the size of all memory regions, the size of the allocated array of the memory
regions and pointer to the array of the memblock_region structures. memblock_region is a structure which describes a
Memblock
240
Linux Inside
struct memblock_region {
phys_addr_t base;
phys_addr_t size;
unsigned long flags;
#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
int nid;
#endif
};
memblock_region provides base address and size of the memory region, flags which can be:
Also memblock_region provides integer field - numa node selector, if the CONFIG_HAVE_MEMBLOCK_NODE_MAP configuration option
is enabled.
Schematically we can imagine it as:
+---------------------------+ +---------------------------+
| memblock | | |
| _______________________ | | |
| | memory | | | Array of the |
| | memblock_type |-|-->| membock_region |
| |_______________________| | | |
| | +---------------------------+
| _______________________ | +---------------------------+
| | reserved | | | |
| | memblock_type |-|-->| Array of the |
| |_______________________| | | memblock_region |
| | | |
+---------------------------+ +---------------------------+
These three structures: memblock , memblock_type and memblock_region are main in the Memblock . Now we know about it
and can look at Memblock initialization process.
Memblock initialization
As all API of the memblock described in the include/linux/memblock.h header file, all implementation of these function is in
the mm/memblock.c source code file. Let's look at the top of the source code file and we will see the initialization of the
memblock structure:
Memblock
241
Linux Inside
.current_limit = MEMBLOCK_ALLOC_ANYWHERE,
};
Here we can see initialization of the memblock structure which has the same name as structure - memblock . First of all note
on __initdata_memblock . Defenition of this macro looks like:
#ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
#define __init_memblock __meminit
#define __initdata_memblock __meminitdata
#else
#define __init_memblock
#define __initdata_memblock
#endif
You can note that it depends on CONFIG_ARCH_DISCARD_MEMBLOCK . If this configuration option is enabled, memblock code will
be put to the .init section and it will be released after the kernel is booted up.
Next we can see initialization of the memblock_type memory , memblock_type reserved and memblock_type physmem fields of the
memblock structure. Here we are interested only in the memblock_type.regions initialization process. Note that every
memblock_type field initialized by the arrays of the memblock_region :
Every array contains 128 memory regions. We can see it in the INIT_MEMBLOCK_REGIONS macro definition:
Note that all arrays are also defined with the __initdata_memblock macro which we already saw in the memblock strucutre
initialization (read above if you've forgot).
The last two fields describe that bottom_up allocation is disabled and the limit of the current Memblock is:
which is 0xffffffffffffffff .
On this step initialization of the memblock structure finished and we can look on the Memblock API.
Memblock API
Ok we have finished with initilization of the memblock structure and now we can look on the Memblock API and its
implementation. As I said above, all implementation of the memblock presented in the mm/memblock.c. To understand how
memblock works and is implemented, let's look at its usage first of all. There are a couple of places in the linux kernel where
memblock is used. For example let's take memblock_x86_fill function from the arch/x86/kernel/e820.c. This function goes
through the memory map provided by the e820 and adds memory regions reserved by the kernel to the memblock with the
memblock_add function. As we met memblock_add function first, let's start from it.
This function takes physical base address and size of the memory region and adds it to the memblock . memblock_add
Memblock
242
Linux Inside
function does not do anything special in its body, but just calls:
function. We pass memory block type - memory , physical base address and size of the memory region, maximum number
of nodes which are zero if CONFIG_NODES_SHIFT is not set in the configuration file or CONFIG_NODES_SHIFT if it is set, and flags.
The memblock_add_range function adds new memory region to the memory block. It starts by checking the size of the given
region and if it is zero it just returns. After this, memblock_add_range checks for existence of the memory regions in the
memblock structure with the given memblock_type . If there are no memory regions, we just fill new memory_region with the
given values and return (we already saw the implementation of this in the First touch of the linux kernel memory manager
framework). If memblock_type is not empty, we start to add new memory region to the memblock with the given
memblock_type .
First of all we get the end of the memory region with the:
memblock_cap_size adjusts size that base + size will not overflow. Its implementation is pretty easy:
memblock_cap_size returns new size which is the smallest value between the given size and ULLONG_MAX - base .
After that we have the end address of the new memory region, memblock_add_range checks overlap and merge conditions
with already added memory regions. Insertion of the new memory region to the memblcok consists of two steps:
Adding of non-overlapping parts of the new memory area as separate regions;
Merging of all neighbouring regions.
We are going through all the already stored memory regions and checking for overlap with the new region:
If the new memory region does not overlap regions which are already stored in the memblock , insert this region into the
memblock with and this is first step, we check that new region can fit into the memory block and call memblock_double_array
in other way:
Memblock
243
Linux Inside
return -ENOMEM;
insert = true;
goto repeat;
memblock_double_array doubles the size of the given regions array. Then we set insert to true and go to the repeat
label. In the second step, starting from the repeat label we go through the same loop and insert the current memory region
into the memory block with the memblock_insert_region function:
As we set insert to true in the first step, now memblock_insert_region will be called. memblock_insert_region has almost
the same implementation that we saw when we insert new region to the empty memblock_type (see above). This function
gets the last memory region:
After this fills memblock_region fields of the new memory region base, size and etc... and increase size of the
memblock_type . In the end of the execution, memblock_add_range calls memblock_merge_regions which merges neighboring
0 0x1000
+-----------------------+
| |
| |
| region1 |
| |
| |
+-----------------------+
And now we want to add region2 to the memblock with the following base address and size:
0x100 0x2000
+-----------------------+
| |
| |
| region2 |
| |
| |
+-----------------------+
In this case set the base address of the new memory region as the end address of the overlapped region with:
Memblock
244
Linux Inside
So it will be 0x1000 in our case. And insert it as we did it already in the second step with:
In this case we insert overlapping portion (we insert only the higher portion, because the lower portion is already in the
overlapped memory region), then the remaining portion and merge these portions with memblock_merge_regions . As I said
above memblock_merge_regions function merges neighboring compatible regions. It goes through the all memory regions
from the given memblock_type , takes two neighboring memory regions - type->regions[i] and type->regions[i + 1] and
checks that these regions have the same flags, belong to the same node and that end address of the first regions is not
equal to the base address of the second region:
If none of these conditions are not true, we update the size of the first region with the size of the next region:
this->size += next->size;
As we update the size of the first memory region with the size of the next memory region, we copy every (in the loop)
memory region which is after the current ( this ) memory region to the one index ago with the memmove function:
And decrease the count of the memory regions which are belongs to the memblock_type :
type->cnt--;
After this we will get two memory regions merged into one:
0 0x2000
+------------------------------------------------+
| |
| |
| region1 |
| |
| |
+------------------------------------------------+
Memblock
245
Linux Inside
That's all. This is the whole principle of the work of the memblock_add_range function.
There is also memblock_reserve function which does the same as memblock_add , but only with one difference. It stores
memblock_type.reserved in the memblock instead of memblock_type.memory .
Of course this is not the full API. Memblock provides an API for not only adding memory and reserved memory regions, but
also:
memblock_remove - removes memory region from memblock;
memblock_find_in_range - finds free area in given range;
memblock_free - releases memory region in memblock;
for_each_mem_range - iterates through memblock areas.
and many more....
First of all this function checks that memblock contains reserved memory regions. If memblock does not contain reserved
memory regions we just return zero. Otherwise we write the physical address of the reserved memory regions array to the
given address and return aligned size of the allocated array. Note that there is PAGE_ALIGN macro used for align. Actually it
depends on size of page:
Implementation of the get_allocated_memblock_memory_regions_info function is the same. It has only one difference,
memblock_type.memory used instead of memblock_type.reserved .
Memblock debugging
There are many calls to memblock_dbg in the memblock implementation. If you pass the memblock=debug option to the kernel
command line, this function will be called. Actually memblock_dbg is just a macro which expands to printk :
Memblock
246
Linux Inside
For example you can see a call of this macro in the memblock_reserve function:
Memblock has also support in debugfs. If you run kernel not in X86 architecture you can access:
/sys/kernel/debug/memblock/memory
/sys/kernel/debug/memblock/reserved
/sys/kernel/debug/memblock/physmem
for getting dump of the memblock contents.
Conclusion
This is the end of the first part about linux kernel memory management. If you have questions or suggestions, ping me on
twitter 0xAX, drop me an email or just create an issue.
Please note that English is not my first language and I am really sorry for any inconvenience. If you found any
mistakes please send me a PR to linux-internals.
Links
e820
numa
debugfs
First touch of the linux kernel memory manager framework
Memblock
247
Linux Inside
be a linear address minus __START_KERNEL_map . Each fix-mapped address maps one page frame and the kernel uses them
as pointers that never change their address. That is the main point of these addresses. As the comment says: to have a
constant address at compile time, but to set the physical address only in the boot process . You can remember that in the
NEXT_PAGE(level2_fixmap_pgt)
.fill 506,8,0
.quad level1_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
.fill 5,8,0
NEXT_PAGE(level1_fixmap_pgt)
.fill 512,8,0
As you can see level2_fixmap_pgt is right after the level2_kernel_pgt which is kernel code+data+bss. Every fix-mapped
address is represented by an integer index which is defined in the fixed_addresses enum from the
arch/x86/include/asm/fixmap.h. For example it contains entries for VSYSCALL_PAGE - if emulation of legacy vsyscall page is
enabled, FIX_APIC_BASE for local apic and etc... In a virtual memory fix-mapped area is placed in the modules area:
+-----------+-----------------+---------------+------------------+
| | | | |
|kernel text| kernel | | vsyscalls |
| mapping | text | Modules | fix-mapped |
|from phys 0| data | | addresses |
| | | | |
+-----------+-----------------+---------------+------------------+
__START_KERNEL_map __START_KERNEL MODULES_VADDR 0xffffffffffffffff
Base virtual address and size of the fix-mapped area are presented by the two following macro:
Here __end_of_permanent_fixed_addresses is an element of the fixed_addresses enum and as I wrote above: Every fixmapped address is represented by an integer index which is defined in the fixed_addresses . PAGE_SHIFT determines size of
a page. For example size of the one page we can get with the 1 << PAGE_SHIFT . In our case we need to get the size of the
fix-mapped area, but not only of one page, that's why we are using __end_of_permanent_fixed_addresses for getting the size
of the fix-mapped area. In my case it's a little more than 536 killobytes. In your case it might be a different number,
because the size depends on amount of the fix-mapped addresses which are depends on your kernel's configuration.
The second FIXADDR_START macro just extracts from the last address of the fix-mapped area its size for getting base virtual
address of the fix-mapped area. FIXADDR_TOP is rounded up address from the base address of the vsyscall space:
The fixed_addresses enums are used as an index to get the virtual address using the fix_to_virt function.
248
Linux Inside
first of all it checks that the index given for the fixed_addresses enum is not greater or equal than __end_of_fixed_addresses
with the BUILD_BUG_ON macro and then returns the result of the __fix_to_virt macro:
Here we shift left the given fix-mapped address index on the PAGE_SHIFT which determines size of a page as I wrote above
and subtract it from the FIXADDR_TOP which is the highest address of the fix-mapped area. There is an inverse function for
getting fix-mapped address from a virtual address:
virt_to_fix takes virtual address, checks that this address is between FIXADDR_START and FIXADDR_TOP and calls
__virt_to_fix macro which implemented as:
A PFN is simply an index within physical memory that is counted in page-sized units. PFN for a physical address could be
trivially defined as (page_phys_addr >> PAGE_SHIFT);
__virt_to_fix clears the first 12 bits in the given address, subtracts it from the last address the of fix-mapped area
( FIXADDR_TOP ) and shifts right result on PAGE_SHIFT which is 12 . Let me explain how it works. As I already wrote we will
clear the first 12 bits in the given address with x & PAGE_MASK . As we subtract this from the FIXADDR_TOP , we will get the last
12 bits of the FIXADDR_TOP which are present. We know that the first 12 bits of the virtual address represent the offset in the
page frame. With the shiting it on PAGE_SHIFT we will get Page frame number which is just all bits in a virtual address besides
the first 12 offset bits. Fix-mapped addresses are used in different places in the linux kernel. IDT descriptor stored there,
Intel Trusted Execution Technology UUID stored in the fix-mapped area started from FIX_TBOOT_BASE index, Xen bootmap
and many more... We already saw a little about fix-mapped addresses in the fifth part about linux kernel initialization. We
used fix-mapped area in the early ioremap initialization. Let's look on it and try to understand what is it ioremap , how it is
implemented in the kernel and how it is releated to the fix-mapped addresses.
ioremap
Linux kernel provides many different primitives to manage memory. For this moment we will touch I/O memory . Every
device is controlled by reading/writing from/to its registers. For example a driver can turn off/on a device by writing to its
registers or get the state of a device by reading from its registers. Besides registers, many devices have buffers where a
driver can write something or read from there. As we know for this moment there are two ways to access device's registers
and data buffers:
through the I/O ports;
Fixmaps and ioremap
249
Linux Inside
$ cat /proc/ioports
0000-0cf7 : PCI Bus 0000:00
0000-001f : dma1
0020-0021 : pic1
0040-0043 : timer0
0050-0053 : timer1
0060-0060 : keyboard
0064-0064 : keyboard
0070-0077 : rtc0
0080-008f : dma page reg
00a0-00a1 : pic2
00c0-00df : dma2
00f0-00ff : fpu
00f0-00f0 : PNP0C04:00
03c0-03df : vesafb
03f8-03ff : serial
04d0-04d1 : pnp 00:06
0800-087f : pnp 00:01
0a00-0a0f : pnp 00:04
0a20-0a2f : pnp 00:04
0a30-0a3f : pnp 00:04
0cf8-0cff : PCI conf1
0d00-ffff : PCI Bus 0000:00
...
...
...
/proc/ioporst provides information about what driver used address of a I/O ports region. All of these memory regions, for
example 0000-0cf7 , were claimed with the request_region function from the include/linux/ioport.h. Actually request_region
is a macro which defied as:
request_region allocates I/O port region. Very often check_region function called before the request_region to check that
the given address range is available and release_region to release memory region. request_region returns pointer to the
resource structure. resource structure presents abstraction for a tree-like subset of system resources. We already saw
resource structure in the firth part about kernel initialization process and it looks as:
struct resource {
resource_size_t start;
resource_size_t end;
const char *name;
unsigned long flags;
struct resource *parent, *sibling, *child;
};
and contains start and end addresses of the resource, name and etc... Every resource structure contains pointers to the
parent , slibling and child resources. As it has parent and childs, it means that every subset of resuorces has root
250
Linux Inside
As I wrote about request_regions is used for registering of I/O port region and this macro used in many places in the
kernel. For example let's look at drivers/char/rtc.c. This source code file provides Real Time Clock interface in the linux
kernel. As every kernel module, rtc module contains module_init definition:
module_init(rtc_init);
where rtc_init is rtc initialization function. This function defined in the same rtc.c source code file. In the rtc_init
function we can see a couple calls of the rtc_request_region functions, which wrap request_region for example:
r = rtc_request_region(RTC_IO_EXTENT);
Here RTC_IO_EXTENT is a size of memory region and it is 0x8 , "rtc" is a name of region and RTC_PORT is:
So with the request_region(RTC_PORT(0), size, "rtc") we register memory region, started at 0x70 and with size 0x8 . Let's
look on the /proc/ioports :
So, we got it! Ok, it was ports. The second way is use of I/O memory. As I wrote above this way is mapping of control
registers and memory of a device to the memory address space. I/O memory is a set of contiguous addresses which are
provided by a device to CPU through a bus. All memory-mapped I/O addresses are not used by the kernel directly. There is
a special ioremap function which allows us to covert the physical address on a bus to the kernel virtual address or in
another words ioremap maps I/O physical memory region to access it from the kernel. The ioremap function takes two
251
Linux Inside
parameters:
start of the memory region;
size of the memory region;
I/O memory mapping API provides function for the checking, requesting and release of a memory region as this does I/O
ports API. There are three functions for it:
request_mem_region
release_mem_region
check_mem_region
Part of these addresses is from the call of the e820_reserve_resources function. We can find call of this function in the
arch/x86/kernel/setup.c and the function itself defined in the arch/x86/kernel/e820.c. e820_reserve_resources goes through
the e820 map and inserts memory regions to the root iomem resource structure. All e820 memory regions which are will be
inserted to the iomem resource will have following types:
252
Linux Inside
paging_init . We do not know anything about vmalloc for now, so let's consider early initialization of the ioremap . First of
all early_ioremap_init checks that fixmap is aligned on page middle directory boundary:
more about BUILD_BUG_ON you can read in the first part about Linux Kernel initialization. So BUILD_BUG_ON macro raises
compilation error if the given expression is true. In the next step after this check, we can see call of the
early_ioremap_setup function from the mm/early_ioremap.c. This function presents generic initialization of the ioremap .
early_ioremap_setup function fills the slot_virt array with the virtual addresses of the early fixmaps. All early fixmaps are
after __end_of_permanent_fixed_addresses in memory. They are stats from the FIX_BITMAP_BEGIN (top) and ends with
FIX_BITMAP_END (down). Actually there are 512 temporary boot-time mappings, used by early ioremap :
#define NR_FIX_BTMAPS 64
#define FIX_BTMAPS_SLOTS 8
#define TOTAL_FIX_BTMAPS (NR_FIX_BTMAPS * FIX_BTMAPS_SLOTS)
and early_ioremap_setup :
the slot_virt and other arrays are defined in the same source code file:
slot_virt contains virtual addresses of the fix-mapped areas, prev_map array contains addresses of the early ioremap
areas. Note that I wrote above: Actually there are 512 temporary boot-time mappings, used by early ioremap and you can
see that all arrays defined with the __initdata attribute which means that this memory will be released after kernel
initialization process. After early_ioremap_setup finished to work, we're getting page middle directory where early ioremap
beginning with the early_ioremap_pmd function which just gets the base address of the page global directory and calculates
the page middle directory for the given address:
After this we fills bm_pte (early ioremap page table entries) with zeros and call the pmd_populate_kernel function:
253
Linux Inside
pmd = early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN));
memset(bm_pte, 0, sizeof(bm_pte));
pmd_populate_kernel(&init_mm, pmd, bm_pte);
init_mm - memory descriptor of the init process (you can read about it in the previous part);
pmd - page middle directory of the beginning of the ioremap fixmaps;
bm_pte - early ioremap page table entries array which defined as:
The pmd_popularte_kernel function defined in the arch/x86/include/asm/pgalloc.h and populates given page middle
directory ( pmd ) with the given page table entries ( bm_pte ):
That's all. Early ioremap is ready to use. There are a couple of checks in the early_ioremap_init function, but they are not
so important, anyway initialization of the ioremap is finished.
254
Linux Inside
First of all in the __early_ioremap , we goes through the all early ioremap fixmap slots and check first free are in the
prev_map array and remember it's number in the slot variable and set up size as we found it:
slot = -1;
for (i = 0; i < FIX_BTMAPS_SLOTS; i++) {
if (!prev_map[i]) {
slot = i;
break;
}
}
...
...
...
prev_size[slot] = size;
last_addr = phys_addr + size - 1;
Here we are using PAGE_MASK for clearing all bits in the phys_addr besides first 12 bits. PAGE_MASK macro defined as:
We know that size of a page is 4096 bytes or 1000000000000 in binary. PAGE_SIZE - 1 will be 111111111111 , but with ~ , we
will get 000000000000 , but as we use ~PAGE_MASK we will get 111111111111 again. On the second line we do the same but
clear first 12 bits and getting page-aligned size of the area on the third line. We getting aligned area and now we need to
get the number of pages which are occupied by the new ioremap are and calculate the fix-mapped index from
fixed_addresses in the next steps:
Now we can fill fix-mapped area with the given physical addresses. Every iteration in the loop, we call __early_set_fixmap
function from the arch/x86/mm/ioremap.c, increase given physical address on page size which is 4096 bytes and update
addresses index and number of pages:
The __early_set_fixmap function gets the page table entry (stored in the bm_pte , see above) for the given physical address
with:
pte = early_ioremap_pte(addr);
255
Linux Inside
In the next step of the early_ioremap_pte we check the given page flags with the pgprot_val macro and calls set_pte or
pte_clear depends on it:
if (pgprot_val(flags))
set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
else
pte_clear(&init_mm, addr, pte);
As you can see above, we passed FIXMAP_PAGE_IO as flags to the __early_ioremap . FIXMPA_PAGE_IO expands to the:
(__PAGE_KERNEL_EXEC | _PAGE_NX)
flags, so we call set_pte function for setting page table entry which works in the same manner as set_pmd but for PTEs
(read above about it). As we set all PTEs in the loop, we can see the call of the __flush_tlb_one function:
__flush_tlb_one(addr);
This function defined in the arch/x86/include/asm/tlbflush.h and calls __flush_tlb_single or __flush_tlb depends on value
of the cpu_has_invlpg :
__flush_tlb_one function invalidates given address in the TLB. As you just saw we updated paging structure, but TLB not
informed of changes, that's why we need to do it manually. There are two ways how to do it. First is update cr3 control
register and __flush_tlb function does this:
native_write_cr3(native_read_cr3());
The second method is to use invlpg instruction invalidates TLB entry. Let's look on __flush_tlb_one implementation. As
you can see first of all it checks cpu_has_invlpg which defined as:
If a CPU support invlpg instruction, we call the __flush_tlb_single macro which expands to the call of the
__native_flush_tlb_single :
256
Linux Inside
or call __flush_tlb which just updates cr3 register as we saw it above. After this step execution of the __early_set_fixmap
function is finsihed and we can back to the __early_ioremap implementation. As we set fixmap area for the given address,
need to save the base virtual address of the I/O Re-mapped area in the prev_map with the slot index:
prev_map[slot] = NULL;
That's all about fixmaps and ioremap . Of course this part does not cover full features of the ioremap , it was only early
ioremap, but there is also normal ioremap. But we need to know more things than now before it.
So, this is the end!
Conclusion
This is the end of the second part about linux kernel memory management. If you have questions or suggestions, ping me
on twitter 0xAX, drop me an email or just create an issue.
Please note that English is not my first language and I am really sorry for any inconvenience. If you found any
mistakes please send me a PR to linux-internals.
Links
apic
vsyscall
Intel Trusted Execution Technology
Xen
Real Time Clock
e820
Memory management unit
TLB
Paging
Linux kernel memory management Part 1.
257
Linux Inside
Concepts
258
Linux Inside
Per-CPU variables
Per-CPU variables are one of the kernel features. You can understand what this feature means by reading its name. We
can create a variable and each processor core will have its own copy of this variable. We take a closer look on this feature
and try to understand how it is implemented and how it works in this part.
The kernel provides API for creating per-cpu variables - DEFINE_PER_CPU macro:
This macro defined in the include/linux/percpu-defs.h as many other macros for work with per-cpu variables. Now we will
see how this feature is implemented.
Take a look at the DECLARE_PER_CPU definition. We see that it takes 2 parameters: type and name , so we can use it to
create per-cpu variable, for example like this:
DEFINE_PER_CPU(int, per_cpu_n)
We pass the type and the name of our variable. DEFI_PER_CPU calls DEFINE_PER_CPU_SECTION macro and passes the same
two paramaters and empty string to it. Let's look at the definition of the DEFINE_PER_CPU_SECTION :
#define __PCPU_ATTRS(sec) \
__percpu __attribute__((section(PER_CPU_BASE_SECTION sec))) \
PER_CPU_ATTRIBUTES
After all macros are expanded we will get global per-cpu variable:
It means that we will have per_cpu_n variable in the .data..percpu section. We can find this section in the vmlinux :
Ok, now we know that when we use DEFINE_PER_CPU macro, per-cpu variable in the .data..percpu section will be created.
When the kernel initilizes it calls setup_per_cpu_areas function which loads .data..percpu section multiply times, one
section per CPU.
Per-CPU variables
259
Linux Inside
Let's look on the per-CPU areas initialization process. It start in the init/main.c from the call of the setup_per_cpu_areas
function which defined in the arch/x86/kernel/setup_percpu.c.
The setup_per_cpu_areas starts from the output information about the Maximum number of CPUs set during kernel
configuration with CONFIG_NR_CPUS configuration option, actual number of CPUs, nr_cpumask_bits is the same that NR_CPUS
bit for the new cpumask operators and number of NUMA nodes.
We can see this output in the dmesg:
In the next step we check percpu first chunk allocator. All percpu areas are allocated in chunks. First chunk is used for the
static percpu variables. Linux kernel has percpu_alloc command line parameters which provides type of the first chunk
allocator. We can read about it in the kernel documentation:
early_param("percpu_alloc", percpu_alloc_setup);
Where percpu_alloc_setup function sets the pcpu_chosen_fc variable depends on the percpu_alloc parameter value. By
default first chunk allocator is auto :
If percpu_alooc parameter not given to the kernel command line, the embed allocator will be used wich as you can
understand embed the first percpu chunk into bootmem with the memblock. The last allocator is first chunk page allocator
which maps first chunk with PAGE_SIZE pages.
As I wrote about first of all we make a check of the first chunk allocator type in the setup_per_cpu_areas . First of all we
check that first chunk allocator is not page:
if (pcpu_chosen_fc != PCPU_FC_PAGE) {
...
...
...
}
If it is not PCPU_FC_PAGE , we will use embed allocator and allocate space for the first chunk with the pcpu_embed_first_chunk
function:
Per-CPU variables
260
Linux Inside
rc = pcpu_embed_first_chunk(PERCPU_FIRST_CHUNK_RESERVE,
dyn_size, atom_size,
pcpu_cpu_distance,
pcpu_fc_alloc, pcpu_fc_free);
As I wrote above, the pcpu_embed_first_chunk function embeds the first percpu chunk into bootmem. As you can see we
pass a couple of parameters to the pcup_embed_first_chunk , they are
PERCPU_FIRST_CHUNK_RESERVE - the size of the reserved space for the static percpu variables;
dyn_size - minimum free size for dynamic allocation in byte;
atom_size - all allocations are whole multiples of this and aligned to this parameter;
pcpu_cpu_distance - callback to determine distance between cpus;
pcpu_fc_alloc - function to allocate percpu page;
pcpu_fc_free - function to release percpu page.
If first chunk allocator is PCPU_FC_PAGE , we will use the pcpu_page_first_chunk instead of the pcpu_embed_first_chunk . After
that percpu areas up, we setup percpu offset and its segment for the every CPU with the setup_percpu_segment function
(only for x86 systems) and move some early data from the arrays to the percpu variables ( x86_cpu_to_apicid ,
irq_stack_ptr and etc...). After the kernel finished the initialization process, we have loaded N .data..percpu sections,
where N is the number of CPU, and section used by bootstrap processor will contain uninitialized variable created with
DEFINE_PER_CPU macro.
#define get_cpu_var(var) \
(*({ \
preempt_disable(); \
this_cpu_ptr(&var); \
}))
Linux kernel is preemptible and accessing a per-cpu variable requires to know which processor kernel running on. So,
current code must not be preempted and moved to the another CPU while accessing a per-cpu variable. That's why first of
all we can see call of the preempt_disable function. After this we can see call of the this_cpu_ptr macro, which looks as:
and
Per-CPU variables
261
Linux Inside
where per_cpu_ptr returns a pointer to the per-cpu variable for the given cpu (second parameter). After that we got per-cpu
variables and made any manipulations on it, we must call put_cpu_var macro which enables preemption with call of
preempt_enable function. So the typical usage of a per-cpu variable is following:
get_cpu_var(var);
...
//Do something with the 'var'
...
put_cpu_var(var);
As I wrote above, this macro returns per-cpu variable for the given cpu. First of all it calls __verify_pcpu_ptr :
#define __verify_pcpu_ptr(ptr)
do {
const void __percpu *__vpp_verify = (typeof((ptr) + 0))NULL;
(void)__vpp_verify;
} while (0)
where NR_CPUS is the number of CPUs. __per_cpu_offset array filled with the distances between cpu-variables copies. For
example all per-cpu data is X bytes size, so if we access __per_cpu_offset[Y] , so X*Y will be accessed. Let's look at the
SHIFT_PERCPU_PTR implementation:
RELOC_HIDE just returns offset (typeof(ptr)) (__ptr + (off)) and it will be pointer of the variable.
That's all! Of course it is not the full API, but the general part. It can be hard for the start, but to understand per-cpu
variables feature need to understand mainly include/linux/percpu-defs.h magic.
Per-CPU variables
262
Linux Inside
When per_cpu_ptr called for example for getting pointer on the certain per-cpu variable for the third CPU,
__per_cpu_offset array will be accessed, where every index points to the certain CPU.
That's all.
Per-CPU variables
263
Linux Inside
CPU masks
Introduction
Cpumasks is a special way provided by the Linux kernel to store information about CPUs in the system. The relevant source
code and header files which are contains API for Cpumasks manipulating:
include/linux/cpumask.h
lib/cpumask.c
kernel/cpu.c
As comment says from the include/linux/cpumask.h: Cpumasks provide a bitmap suitable for representing the set of CPU's
in a system, one bit position per CPU number. We already saw a bit about cpumask in the boot_cpu_init function from the
Kernel entry point part. This function makes first boot cpu online, active and etc...:
set_cpu_online(cpu, true);
set_cpu_active(cpu, true);
set_cpu_present(cpu, true);
set_cpu_possible(cpu, true);
set_cpu_possible is a set of cpu ID's which can be plugged in anytime during the life of that system boot. cpu_present
represents which CPUs are currently plugged in. cpu_online represents subset of the cpu_present and indicates CPUs
which are available for scheduling. These masks depends on CONFIG_HOTPLUG_CPU configuration option and if this option is
disabled possible == present and active == online . Implementation of the all of these functions are very similar. Every
function checks the second parameter. If it is true , calls cpumask_set_cpu or cpumask_clear_cpu otherwise.
There are two ways for a cpumask creation. First is to use cpumask_t . It defined as:
It wraps cpumask structure which contains one bitmak bits field. DECLARE_BITMAP macro gets two parameters:
bitmap name;
number of bits.
and creates an array of unsigned long with the give name. It's implementation is pretty easy:
#define DECLARE_BITMAP(name,bits) \
unsigned long name[BITS_TO_LONGS(bits)]
where BITS_TO_LONG :
As we learning x86_64 architecture, unsigned long is 8-bytes size and our array will contain only one element:
Cpumasks
264
Linux Inside
NR_CPUS macro presents the number of the CPUs in the system and depends on the CONFIG_NR_CPUS macro which defined
#ifndef CONFIG_NR_CPUS
#define CONFIG_NR_CPUS 1
#endif
#define NR_CPUS CONFIG_NR_CPUS
The second way to define cpumask is to use DECLARE_BITMAP macro directly and to_cpumask macro which convertes given
bitmap to the struct cpumask * :
#define to_cpumask(bitmap) \
((struct cpumask *)(1 ? (bitmap) \
: (void *)sizeof(__check_is_bitmap(bitmap))))
We can see ternary operator operator here which is true every time. __check_is_bitmap inline function defined as:
And returns 1 every time. We need in it here only for one purpose: In compile time it checks that given bitmap is a bitmap,
or with another words it checks that given bitmap has type - unsigned long * . So we just pass cpu_possible_bits to the
to_cpumask macro for converting array of unsigned long to the struct cpumask * .
cpumask API
As we can define cpumask with one of the method, Linux kernel provides API for manipulating a cpumask. Let's consider
one of the function which presented above. For example set_cpu_online . This function takes two parameters:
Number of CPU;
CPU status;
Implementation of this function looks as:
First of all it checks the second state parameter and calls cpumask_set_cpu or cpumask_clear_cpu depends on it. Here we
can see casting to the struct cpumask * of the second parameter in the cpumask_set_cpu . In our case it is cpu_online_bits
which is bitmap and defined as:
Cpumasks
265
Linux Inside
cpumask_set_cpu function makes only one call of the set_bit function inside:
set_bit function takes two parameter too, and sets a given bit (first parameter) in the memory (second parameter or
cpu_online_bits bitmap). We can see here that before set_bit will be called, its two parameter will be passed to the
cpumask_check;
cpumask_bits.
Let's consider these two macro. First if cpumask_check does nothing in our case and just returns given parameter. The
second cpumask_bits just returns bits field from the given struct cpumask * structure:
This function looks scarry, but it is not so hard as it seems. First of all it passes nr or number of the bit to the IS_IMMEDIATE
macro which just makes call of the GCC internal __builtin_constant_p function:
__builtin_constant_p checks that given parameter is known constant at compile-time. As our cpu is not compile-time
executed. This allows to synchronize memory access, preventing simultaneous access of multiple processors (or devices DMA controller for example) to one memory cell.
BITOP_ADDR casts given parameter to the (*(volatile long *) and adds +m constraints. + means that this operand is bot
read and written by the instruction. m shows that this is memory operand. BITOP_ADDR is defined as:
Cpumasks
266
Linux Inside
Next is the memory clobber. It tells the compiler that the assembly code performs memory reads or writes to items other
than those listed in the input and output operands (for example, accessing the memory pointed to by one of the input
parameters).
Ir - immideate register operand.
bts instruction sets given bit in a bit string and stores the value of a given bit in the CF flag. So we passed cpu number
which is zero in our case and after set_bit will be executed, it sets zero bit in the cpu_online_bits cpumask. It would
mean that the first cpu is online at this moment.
Besides the set_cpu_* API, cpumask ofcourse provides another API for cpumasks manipulation. Let's consider it in shoft.
This macro returns amount of the online CPUs. It calls cpumask_weight function with the cpu_online_mask bitmap (read
about about it). cpumask_wieght function makes an one call of the bitmap_wiegt function with two parameters:
cpumask bitmap;
nr_cpumask_bits - which is NR_CPUS in our case.
and calculates amount of the bits in the given bitmap. Besides the num_online_cpus , cpumask provides macros for the all
CPU states:
num_possible_cpus;
num_active_cpus;
cpu_online;
cpu_possible.
and many more.
Besides that Linux kernel provides following API for the manipulating of cpumask :
for_each_cpu - iterates over every cpu in a mask;
for_each_cpu_not - iterates over every cpu in a complemented mask;
cpumask_clear_cpu - clears a cpu in a cpumask;
cpumask_test_cpu - tests a cpu in a mask;
cpumask_setall - set all cpus in a mask;
cpumask_size - returns size to allocate for a 'struct cpumask' in bytes;
Cpumasks
267
Linux Inside
Links
cpumask documentation
Cpumasks
268
Linux Inside
269
Linux Inside
struct list_head {
struct list_head *next, *prev;
};
You can note that it is different from many lists implementations which you could see. For example this doubly linked list
structure from the glib:
struct GList {
gpointer data;
GList *next;
GList *prev;
};
Usually a linked list structure contains a pointer to the item. Linux kernel implementation of the list does not. So the main
question is - where does the list store the data? . The actual implementation of lists in the kernel is - Intrusive list . An
intrusive linked list does not contain data in its nodes - A node just contains pointers to the next and previous node and list
nodes part of the data that are added to the list. This makes the data structure generic, so it does not care about entry data
type anymore.
For example:
struct nmi_desc {
spinlock_t lock;
struct list_head head;
};
Let's look at some examples, how list_head is used in the kernel. As I already wrote about, there are many, really many
different places where lists are used in the kernel. Let's look for example in miscellaneous character drivers. Misc character
drivers API from the drivers/char/misc.c for writing small drivers for handling simple hardware or virtual devices. This drivers
share major number:
#define MISC_MAJOR 10
but has own minor number. For example you can see it with:
ls -l /dev | grep 10
crw------- 1 root root 10, 235 Mar 21 12:01 autofs
drwxr-xr-x 10 root root 200 Mar 21 12:01 cpu
crw------- 1 root root 10, 62 Mar 21 12:01 cpu_dma_latency
crw------- 1 root root 10, 203 Mar 21 12:01 cuse
270
Linux Inside
Now let's look how lists are used in the misc device drivers. First of all let's look on miscdevice structure:
struct miscdevice
{
int minor;
const char *name;
const struct file_operations *fops;
struct list_head list;
struct device *parent;
struct device *this_device;
const char *nodename;
mode_t mode;
};
We can see the fourth field in the miscdevice structure - list which is list of registered devices. In the beginning of the
source code file we can see definition of the:
static LIST_HEAD(misc_list);
#define LIST_HEAD(name) \
struct list_head name = LIST_HEAD_INIT(name)
and initializes it with the LIST_HEAD_INIT macro which set previous and next entries:
Now let's look on the misc_register function which registers a miscellaneous device. At the start it initializes miscdevice>list with the INIT_LIST_HEAD function:
INIT_LIST_HEAD(&misc->list);
271
Linux Inside
In the next step after device created with the device_create function we add it to the miscellaneous devices list with:
list_add(&misc->list, &misc_list);
Kernel list.h provides this API for the addition of new entry to the list. Let's look on it's implementation:
Here we set new item between prev and next . So misc list which we defined at the start with the LIST_HEAD_INIT macro
will contain previous and next pointers to the miscdevice->list .
There is still only one question how to get list's entry. There is special special macro for this point:
After this we can access to the any miscdevice field with p->minor or p->name and etc... Let's look on the list_entry
implementation:
272
Linux Inside
As we can see it just calls container_of macro with the same arguments. For the first look container_of looks strange:
First of all you can note that it consists from two expressions in curly brackets. Compiler will evaluate the whole block in the
curly braces and use the value of the last expression.
For example:
#include <stdio.h>
int main() {
int i = 0;
printf("i = %d\n", ({++i; ++i;}));
return 0;
}
will print 2 .
The next point is typeof , it's simple. As you can understand from its name, it just returns the type of the given variable.
When I first saw the implementation of the container_of macro, the strangest thing for me was the zero in the ((type *)0)
expression. Actually this pointer magic calculates the offset of the given field from the address of the structure, but as we
have 0 here, it will be just a zero offset alongwith the field width. Let's look at a simple example:
#include <stdio.h>
struct s {
int field1;
char field2;
char field3;
};
int main() {
printf("%p\n", &((struct s*)0)->field3);
return 0;
}
Let's summarize all about container_of macro. container_of macro returns address of the structure by the given address
of the structure's field with list_head type, the name of the structure field with list_head type and type of the container
structure. At the first line this macro declares the __mptr pointer which points to the field of the structure that ptr points to
and assigns it to the ptr . Now ptr and __mptr point to the same address. Technically we don't need this line but its
useful for type checking. First line ensures that that given structure ( type parameter) has a member called member . In the
second line it calculates offset of the field from the structure with the offsetof macro and subtracts it from the structure
Doubly linked list
273
Linux Inside
274
Linux Inside
include/linux/radix-tree.h
lib/radix-tree.c
Lets talk about what is radix tree . Radix tree is a compressed trie where trie is a data structure which implements
interface of an associative array and allows to store values as key-value . The keys are usually strings, but any other data
type can be used as well. Trie is different from any n-tree in its nodes. Nodes of a trie do not store keys, instead, a node of
a trie stores single character labels. The key which is related to a given node is derived by traversing from the root of the
tree to this node. For example:
+-----------+
||
|" "|
| |
+------+-----------+------+
||
||
+----v------++-----v-----+
||||
|g||c|
| | | |
+-----------++-----------+
||
||
+----v------++-----v-----+
||||
|o||a|
| | | |
+-----------++-----------+
|
|
+-----v-----+
||
|t|
| |
+-----------+
So in this example, we can see the trie with keys, go and cat . The compressed trie or radix tree differs from trie ,
such that all intermediates nodes which have only one child are removed.
Radix tree in linux kernel is the datastructure which maps values to the integer key. It is represented by the following
structures from the file include/linux/radix-tree.h:
struct radix_tree_root {
unsigned int height;
gfp_t gfp_mask;
struct radix_tree_node __rcu *rnode;
};
This structure presents the root of a radix tree and contains three fields:
Radix tree
275
Linux Inside
struct radix_tree_node {
unsigned int path;
unsigned int count;
union {
struct {
struct radix_tree_node *parent;
void *private_data;
};
struct rcu_head rcu_head;
};
/* For tree user */
struct list_head private_list;
void __rcu *slots[RADIX_TREE_MAP_SIZE];
unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS];
};
This structure contains information about the offset in a parent and height from the bottom, count of the child nodes and
fields for accessing and freeing a node. The fields are described below:
path - offset in parent & height from the bottom;
count - count of the child nodes;
parent - pointer to the parent node;
private_data - used by the user of a tree;
rcu_head - used for freeing a node;
private_list - used by the user of a tree;
The two last fields of the radix_tree_node - tags and slots are important and interesting. Every node can contains a set
of slots which are store pointers to the data. Empty slots in the linux kernel radix tree implementation store NULL . Radix
tree in the linux kernel also supports tags which are associated with the tags fields in the radix_tree_node structure. Tags
allow to set individual bits on records which are stored in the radix tree.
Now we know about radix tree structure, time to look on its API.
RADIX_TREE(name, gfp_mask);
`
As you can see we pass the name parameter, so with the RADIX_TREE macro we can define and initialize radix tree with the
given name. Implementation of the RADIX_TREE is easy:
Radix tree
276
Linux Inside
At the beginning of the RADIX_TREE macro we define instance of the radix_tree_root structure with the given name and call
RADIX_TREE_INIT macro with the given mask. The RADIX_TREE_INIT macro just initializes radix_tree_root structure with the
where:
makes the same initialziation with default values as it does RADIX_TREE_INIT macro.
The next are two functions for the inserting and deleting records to/from a radix tree:
radix_tree_insert ;
radix_tree_delete .
Radix tree
277
Linux Inside
and returns number of records, sorted by the keys, starting from the first index. Number of the returned records will be not
greater than max_items value.
And the last radix_tree_lookup_slot function will return the slot which will contain the data.
Links
Radix tree
Trie
Radix tree
278
Linux Inside
Theory
This chapter describes various theoretical concepts and concepts which are not directly related to practice but useful to
know.
Paging
Elf64 format
Theory
279
Linux Inside
Paging
Introduction
In the fifth part of the series Linux kernel booting process we finished to learn what and how kernel does on the earliest
stage. In the next step kernel will initialize different things like initrd mounting, lockdep initialization, and many many
different things, before we can see how the kernel will run the first init process.
Yeah, there will be many different things, but many many and once again many work with memory.
In my view, memory management is one of the most complex part of the linux kernel and in system programming generally.
So before we will proceed with the kernel initialization stuff, we will get acquainted with the paging.
Paging is a process of translation a linear memory address to a physical address. If you have read previous parts, you can
remember that we saw segmentation in the real mode when physical address calculated by shifting a segment register on
four and adding offset. Or also we saw segmentation in the protected mode, where we used the tables of descriptors and
base addresses from descriptors with offsets to calculate physical addresses. Now we are in 64-bit mode and that we will
see paging.
As Intel manual says:
Paging provides a mechanism for implementing a conventional demand-paged, virtual-memory system where
sections of a programs execution environment are mapped into physical memory as needed.
So... I will try to explain how paging works in theory in this post. Of course it will be closely related with the linux kernel for
x86_64 , but we will not go into deep details (at least in this post).
Enabling paging
There are three paging modes:
32-bit paging;
PAE paging;
IA-32e paging.
We will see explanation only last mode here. To enable IA-32e paging paging mode need to do following things:
set CR0.PG bit;
set CR4.PAE bit;
set IA32_EFER.LME bit.
We already saw setting of this bits in the arch/x86/boot/compressed/head_64.S:
and
Paging
280
Linux Inside
Paging structures
Paging divides the linear address space into fixed-size pages. Pages can be mapped into the physical address space or
even external storage. This fixed size is 4096 bytes for the x86_64 linux kernel. For a linear address translation to a
physical address used special structures. Every structure is 4096 bytes size and contains 512 entries (this only for PAE
and IA32_EFER.LME modes). Paging structures are hierarchical and linux kernel uses 4 level paging for x86_64 . CPU uses a
part of the linear address to identify entry of the another paging structure which is at the lower level or physical memory
region ( page frame ) or physical address in this region ( page offset ). The address of the top level paging structure located
in the cr3 register. We already saw this in the arch/x86/boot/compressed/head_64.S:
We built page table structures and put the address of the top-level structure to the cr3 register. Here cr3 is used to store
the address of the top-level PML4 structure or Page Global Directory as it calls in linux kernel. cr3 is 64-bit register and
has the following structure:
63 52 51 32
-------------------------------------------------------------------------------| | |
| Reserved MBZ | Address of the top level structure |
| | |
-------------------------------------------------------------------------------31 12 11 5 4 3 2 0
-------------------------------------------------------------------------------| | | P | P | |
| Address of the top level structure | Reserved | C | W | Reserved |
| | | D | T | |
--------------------------------------------------------------------------------
paging structure level-3, 29:21 bits stores an index into the paging structure level-2, 20:12 bits stores an index into
the paging structure level-1 and 11:0 bits provide the byte offset into the physical page.
schematically, we can imagine it like this:
Paging
281
Linux Inside
Every access to a linear address is either a supervisor-mode access or a user-mode access. This access determined by
the CPL (current privilege level). If CPL < 3 it is a supervisor mode access level and user mode access level in other ways.
For example top level page table entry contains access bits and has the following structure:
63 62 52 51 32
-------------------------------------------------------------------------------| N | | |
| | Available | Address of the paging structure on lower level |
| X | | |
-------------------------------------------------------------------------------31 12 11 9 8 7 6 5 4 3 2 1 0
-------------------------------------------------------------------------------| | | M |I| | P | P |U|W| |
| Address of the paging structure on lower level | AVL | B |G|A| C | W | | | P |
| | | Z |N| | D | T |S|R| |
--------------------------------------------------------------------------------
Where:
63 bit - N/X bit (No Execute Bit) - presents ability to execute the code from physical pages mapped by the table entry;
62:52 bits - ignored by CPU, used by system software;
51:12 bits - stores physical address of the lower level paging structure;
12:9 bits - ignored by CPU;
MBZ - must be zero bits;
Ignored bits;
A - accessed bit indicates was physical page or page structure accessed;
PWT and PCD used for cache;
U/S - user/supervisor bit controls user access to the all physical pages mapped by this table entry;
R/W - read/write bit controls read/write access to the all physical pages mapped by this table entry;
P - present bit. Current bit indicates was page table or physical page loaded into primary memory or not.
Ok, now we know about paging structures and it's entries. Let's see some details about 4-level paging in linux kernel.
Paging
282
Linux Inside
We can see 0xffffffff81efe497 here. I'm not sure that you have so big RAM. But anyway start_kernel and
x86_64_start_kernel will be executed. The address space in x86_64 is 2^64 size, but it's too large, that's why used smaller
address space, only 48-bits wide. So we have situation when physical address limited with 48 bits, but addressing still
performed with 64 bit pointers. How to solve this problem? Ok, look on the diagram:
0xffffffffffffffff +-----------+
| |
| | Kernelspace
| |
0xffff800000000000 +-----------+
| |
| |
| hole |
| |
| |
0x00007fffffffffff +-----------+
| |
| | Userspace
| |
0x0000000000000000+-----------+
This solution is sign extension . Here we can see that low 48 bits of a virtual address can be used for addressing. Bits
63:48 can be or 0 or 1. Note that all virtual address space is spliten on 2 parts:
Kernel space
Userspace
Userspace occupies the lower part of the virtual address space, from 0x000000000000000 to 0x00007fffffffffff and kernel
space occupies the highest part from the 0xffff8000000000 to 0xffffffffffffffff . Note that bits 63:48 is 0 for userspace
and 1 for kernel space. All addresses which are in kernel space and in userspace or in another words which higher 63:48
bits zero or one calls canonical addresses. There is non-canonical area between these memory regions. Together this two
memory regions (kernel space and user space) are exactly 2^48 bits. We can find virtual memory map with 4 level page
tables in the Documentation/x86/x86_64/mm.txt:
Paging
283
Linux Inside
We can see here memory map for user space, kernel space and non-canonical area between. User space memory map is
simple. Let's take a closer look on the kernel space. We can see that it starts from the guard hole which reserved for
hypervisor. We can find definition of this guard hole in the arch/x86/include/asm/page_64_types.h:
Previously this guard hole and __PAGE_OFFSET was from 0xffff800000000000 to 0xffff80ffffffffff for preventing of access
to non-canonical area, but later was added 3 bits for hypervisor.
Next is the lowest usable address in kernel space - ffff880000000000 . This virtual memory region is for direct mapping of
the all physical memory. After the memory space which mapped all physical address - guard hole, it needs to be between
direct mapping of the all physical memory and vmalloc area. After the virtual memory map for the first terabyte and unused
hole after it, we can see kasan shadow memory. It was added by the commit and provides kernel address sanitizer. After
next unused hole we can se esp fixup stacks (we will talk about it in the other parts) and the start of the kernel text
mapping from the physical address - 0 . We can find definition of this address in the same file as the __PAGE_OFFSET :
Usually kernel's .text start here with the CONFIG_PHYSICAL_START offset. We saw it in the post about ELF64:
Here i checked vmlinux with the CONFIG_PHYSICAL_START is 0x1000000 . So we have the start point of the kernel .text 0xffffffff80000000 and offset - 0x1000000 , the resulted virtual address will be 0xffffffff80000000 + 1000000 =
0xffffffff81000000 .
After the kernel .text region, we can see virtual memory region for kernel modules, vsyscalls and 2 megabytes unused
hole.
We know how looks kernel's virtual memory map and now we can see how a virtual address translates into physical. Let's
take for example following address:
0xffffffff81000000
Paging
284
Linux Inside
That is all. Now you know a little about paging theory and we can go ahead in the kernel source code and see first
initialization steps.
Conclusion
It's the end of this short part about paging theory. Of course this post doesn't cover all details about paging, but soon we will
see it on practice how linux kernel builds paging structures and work with it.
Please note that English is not my first language and I am really sorry for any inconvenience. If you found any
mistakes please send me PR to linux-internals.
Links
Paging on Wikipedia
Intel 64 and IA-32 architectures software developer's manual volume 3A
MMU
ELF64
Documentation/x86/x86_64/mm.txt
Last part - Kernel booting process
Paging
285
Linux Inside
286
Linux Inside
All data stores in a sections in an Elf object file. Sections identified by index in the section header table. Section header
contains following fields:
Section name;
Section type;
Section attributes;
Virtual address in memory;
Offset in file;
Size of section;
Link to other section;
Miscellaneous information;
Address alignment boundary;
Size of entries, if section has table;
And presented with the following elf64_shdr structure in the linux kernel:
And ELF object file also contains other fields/structures which you can find in the Documentation. Better let's look on the
vmlinux .
vmlinux
vmlinux is relocatable ELF object file too. So we can look on it with the readelf util. First of all let's look on a header:
$ readelf -h vmlinux
Elf64
287
Linux Inside
ELF Header:
Magic: 7f 45 4c 46 02 01 01 00 00 00 00 00 00 00 00 00
Class: ELF64
Data: 2's complement, little endian
Version: 1 (current)
OS/ABI: UNIX - System V
ABI Version: 0
Type: EXEC (Executable file)
Machine: Advanced Micro Devices X86-64
Version: 0x1
Entry point address: 0x1000000
Start of program headers: 64 (bytes into file)
Start of section headers: 381608416 (bytes into file)
Flags: 0x0
Size of this header: 64 (bytes)
Size of program headers: 56 (bytes)
Number of program headers: 5
Size of section headers: 64 (bytes)
Number of section headers: 73
Section header string table index: 70
Note that here is address of the startup_64 routine is not ffffffff80000000 , but ffffffff81000000 and now i'll explain why.
We can see following definition in the arch/x86/kernel/vmlinux.lds.S:
. = __START_KERNEL;
...
...
..
/* Text and read-only data */
.text : AT(ADDR(.text) - LOAD_OFFSET) {
_text = .;
...
...
...
}
__START_KERNEL_map is the value from documentation - ffffffff80000000 and __PHYSICAL_START is 0x1000000 . That's why
Elf64
288
Linux Inside
readelf -l vmlinux
Elf file type is EXEC (Executable file)
Entry point 0x1000000
There are 5 program headers, starting at offset 64
Program Headers:
Type Offset VirtAddr PhysAddr
FileSiz MemSiz Flags Align
LOAD 0x0000000000200000 0xffffffff81000000 0x0000000001000000
0x0000000000cfd000 0x0000000000cfd000 R E 200000
LOAD 0x0000000001000000 0xffffffff81e00000 0x0000000001e00000
0x0000000000100000 0x0000000000100000 RW 200000
LOAD 0x0000000001200000 0x0000000000000000 0x0000000001f00000
0x0000000000014d98 0x0000000000014d98 RW 200000
LOAD 0x0000000001315000 0xffffffff81f15000 0x0000000001f15000
0x000000000011d000 0x0000000000279000 RWE 200000
NOTE 0x0000000000b17284 0xffffffff81917284 0x0000000001917284
0x0000000000000024 0x0000000000000024 4
Section to Segment mapping:
Segment Sections...
00 .text .notes __ex_table .rodata __bug_table .pci_fixup .builtin_fw
.tracedata __ksymtab __ksymtab_gpl __kcrctab __kcrctab_gpl
__ksymtab_strings __param __modver
01 .data .vvar
02 .data..percpu
03 .init.text .init.data .x86_cpu_dev.init .altinstructions
.altinstr_replacement .iommu_table .apicdrivers .exit.text
.smp_locks .data_nosave .bss .brk
Here we can see five segments with sections list. All of these sections you can find in the generated linker script at arch/x86/kernel/vmlinux.lds .
That's all. Of course it's not a full description of ELF object format, but if you are interesting in it, you can find documentation
- here
Elf64
289
Linux Inside
Useful links
Linux boot
Linux/x86 boot protocol
Linux kernel parameters
Protected mode
64-ia-32-architectures-software-developer-vol-3a-part-1-manual.pdf
Serial programming
8250 UART Programming
Serial ports on OSDEV
VGA
Video Graphics Array (VGA)
IO
IO port programming
Other architectures
PowerPC and Linux Kernel Inside
Useful links
290
Linux Inside
Contributors
291
Linux Inside
Simarpreet Singh
umatomba
Vaibhav Tulsyan
Brandon Wamboldt
Maxime Leboeuf
Maximilien Richer
marmeladema
Anisse Astier
TheCodeArtist
Ehsun N
Adam Shannon
Donny Nadolny
Ehsun N
Contributors
292